diff --git a/src/blog/2024-05-16-context-matters.mdx b/src/blog/2024-05-16-context-matters.mdx
index a23d9f5a..2906614e 100644
--- a/src/blog/2024-05-16-context-matters.mdx
+++ b/src/blog/2024-05-16-context-matters.mdx
@@ -5,10 +5,11 @@ author: "Tattle"
project: Uli
date: 2024-05-16
tags: machine-learning
-cover: ../images/products/Illustration-context-matters.png
----
-
+
+---
+
+ Building Uli: How the hover-for-metadata feature came about.
In October of 2022, we presented an early version of the Uli browser extension to many people.
By this time we had created a crowdsourced list of slurs in Indian languages. The Uli browser extension looked for these slurs on any webpage that you visited and if it found a match, it would replace that word with a black patch. We called this feature slur redaction.
diff --git a/src/blog/2025-03-07-techdev-uli.mdx b/src/blog/2025-03-07-techdev-uli.mdx
index 7ede2f1a..c22678f8 100644
--- a/src/blog/2025-03-07-techdev-uli.mdx
+++ b/src/blog/2025-03-07-techdev-uli.mdx
@@ -30,22 +30,24 @@ In the process of updating Uli, we also created the following options on the plu
Ahead of workshops we were conducting with college faculty and students, we worked on easing the process of demonstrating the value of Uli to our audience, such as the process of crowdsourcing online, and the metadata display on slur words. There were constraints posed in terms of limited access to laptops, which meant we needed to work out a way to enable the Uli experience on a mobile device as well.
- 1. **Uli Community Page**
+1. **Uli Community Page**
We created a page that would allow users to look at words they added to the crowdsourced list, as well as the contextual information added about it, as below. It was developed using the Phoenix web framework in the Elixir programming language and we use PostgreSQL for our database.
-
+
@@ -26,5 +27,5 @@ Dharini Priscilla, one of the annotators for the Tamil list, also reflected on h
Uli’s plug-in also incorporates ways to contextualise. Users can create their own list of words that they want to see redacted in their feeds. Additionally, since the slur lists are open source, they can be used and contextualised depending on the need. Tattle has also kept an open line of communication, allowing users to write in if they feel a word needs to be removed from the list, and have a compelling reason for the request.
-Illustration: Mitali Panganti
+### Illustration: Mitali Panganti
diff --git a/src/blog/2024-12-05-tattle-mlcommons.mdx b/src/blog/2024-12-05-tattle-mlcommons.mdx
index a47078ed..6cd6ba55 100644
--- a/src/blog/2024-12-05-tattle-mlcommons.mdx
+++ b/src/blog/2024-12-05-tattle-mlcommons.mdx
@@ -8,7 +8,7 @@ tags: responsible-ai
cover: ../images/ml-commons-report-blog-hero.png
---
-## Launch of V1.0 AI Luminate, and how Tattle is involved
+### Launch of V1.0 AI Luminate, and how Tattle is involved
Earlier this year, ML Commons, a global organisation which works to improve AI systems issued an expression of interest for creating prompts in non-English languages.
Tattle was selected as a pilot project to contribute to the benchmark in Hindi, using the participatory approach we followed with Uli [^1], and we commenced working on this project.
diff --git a/src/blog/2025-02-03-hover-feature.mdx b/src/blog/2025-02-03-hover-feature.mdx
index 1896a4e1..24169653 100644
--- a/src/blog/2025-02-03-hover-feature.mdx
+++ b/src/blog/2025-02-03-hover-feature.mdx
@@ -7,8 +7,10 @@ date: 2025-02-03
tags: online-harms
cover: ../images/uli-slur-redaction-image.jpg
---
-
-## Building Uli: How the hover-for-metadata feature came about.
+
+import { Heading } from "grommet";
+
+
+
- 2. **Interactive Input Page**
- For our second workshop, we wanted to have an interactive exercise where we set forth examples of daily life with uncomfortable interactions and problematic words mixed in during conversations.
- Presented with the limitation that students only had access to their mobiles and not laptops, we needed an interface for the students to test it against our database of slurs and our hover feature which would display the metadata behind a slur while highlighting them.
- Therefore we created a separate web page on the Uli website, where students could input the illustrations we had written, and receive an output with the hover feature’s functionality displayed.
- This feature processes the input text, checking it against predefined metadata. Matched words are wrapped with styled HTML elements for highlighting and hover effects, and the result is dynamically rendered.
- React components manage state and lifecycle events for interactions.
+
+2. **Interactive Input Page**
+
+For our second workshop, we wanted to have an interactive exercise where we set forth examples of daily life with uncomfortable interactions and problematic words mixed in during conversations.
+Presented with the limitation that students only had access to their mobiles and not laptops, we needed an interface for the students to test it against our database of slurs and our hover feature which would display the metadata behind a slur while highlighting them.
+Therefore we created a separate web page on the Uli website, where students could input the illustrations we had written, and receive an output with the hover feature’s functionality displayed.
+This feature processes the input text, checking it against predefined metadata. Matched words are wrapped with styled HTML elements for highlighting and hover effects, and the result is dynamically rendered.
+React components manage state and lifecycle events for interactions.
-
+
diff --git a/src/blog/2025-03-12-deepfake-o-meter.mdx b/src/blog/2025-03-12-deepfake-o-meter.mdx
index 0b3a970f..6e3d3a4f 100644
--- a/src/blog/2025-03-12-deepfake-o-meter.mdx
+++ b/src/blog/2025-03-12-deepfake-o-meter.mdx
@@ -21,6 +21,7 @@ Alongside this, we also looked at the different types of deepfakes across variou
As we wanted to assess feasibility, in terms of training/inference and the potential for self-hosting. We did a high level categorisation of all the models in deepfake-o-meter based on attributes such as framework/library used, programming language, and ease of use (easy \- instructions to run the model are given and weights are open, medium \- model needs to be trained, hard \- model needs to be trained or/and training instruction are vague/not available).
+
+
+
+ Read Report
+
+
+
diff --git a/src/blog/misinfocon-india-working-groups.mdx b/src/blog/misinfocon-india-working-groups.mdx
index ee991963..96f795d6 100644
--- a/src/blog/misinfocon-india-working-groups.mdx
+++ b/src/blog/misinfocon-india-working-groups.mdx
@@ -12,7 +12,7 @@ cover: ../images/cover-misinfocon-wg.png
These were the broad topics or working groups that people in the gathering converged on, on day 3:
-# Zombie Claims
+## Zombie Claims
- Research how Zombie Misinformation functions, what it means, how it works and why one time debunking does not work.
- Research wikipedia as a source for zombie misinformation
@@ -25,7 +25,7 @@ These were the broad topics or working groups that people in the gathering conve
- Tell platforms that they can use these systems
- Make this tech available for big and small platforms
-# Reduce Debunking Latency
+### Reduce Debunking Latency
- Work with Meedan’s check tipline to tackle this
- Create popup news rooms around specific events
@@ -35,21 +35,21 @@ These were the broad topics or working groups that people in the gathering conve
- Make this data available to different audiences (newsroom, individuals)
- Solve for claims reappearing in multiple languages by pre-translating every factcheck
-# Platform Failure
+## Platform Failure
- Explore whether self regulation can work. If not then how do you fix it with regulation?
- Compile the tech solutions that we should advocate for.
- Study How do people exploit anonymity to spread misinformation? Can we rely on algorithmic mechanisms for reducing virality from anonymous sources/ platforms?
- Think about tactics that leverage regulatory arbitrage (across geographies).
-# Liberation from Big tech Platforms
+### Liberation from Big tech Platforms
- Create a data donation service
- Create a traceable archive for communities
- Archive for fact checked content (?)
- Build a tool that lets you post on short video platforms while also keeping local archives of content produced.
-# Producing Content
+## Producing Content
- Create a Narrative show that shows how information or misinformation is a part of our daily family life.
Beyond media literacy that is framed as wrong action.
@@ -60,18 +60,18 @@ These were the broad topics or working groups that people in the gathering conve
- Ideate on Digital storytelling for fact checked content beyond long-form text storytelling
- Provide a digest to creators on what topics to tackle
-# Increasing reach of Fact Checks
+### Increasing reach of Fact Checks
- SEO training
- Study and implement UI tweaks that work for different demographics
-# Source Tracking
+## Source Tracking
- Think about citational justice in terms of wikipedia
- Can you use a metadata tag/ poisoning to track networks.
- Libraries and archives are thinking of this in context of provenance and AI
-# MIA Data
+## MIA Data
- A crowdsourced data directory
- Pre-internet Tata used to create a dataset
@@ -79,13 +79,13 @@ These were the broad topics or working groups that people in the gathering conve
- See if we can join the factly project
- Activating networks for maintaining/cleaning datasets
-# Data Void
+## Data Void
- Assist Meedan's work on gendered disinformation:
- Definitions are a problem (one form of a gender based violence)
- Identifying gendered disinformation (support on tech for contextual detection).
-# Critical Reading
+## Critical Reading
- Make Media literacy curriculum content (targeted for age groups for critical thinking)
- Keeping updated based on research
@@ -93,9 +93,9 @@ These were the broad topics or working groups that people in the gathering conve
- Tabulating list of existing curriculum
- Cataloging ‘misinformation techniques’ (informal logics)
-# Social media and archiving
+### Social media and archiving
- Archiving was a theme that came up in multiple working groups. The notes from the dedicated breakout session on the topic can be found [here](https://docs.google.com/document/d/1JwqgALWCH4oKAc0HoTztleW8eWDtL9RK01ul5RX0BHI/edit#heading=h.4cywptwoc66x).
-# More documentation
+## More documentation
The documentation on the work continuing from MisinfoCon is being maitained here: [https://tattle-made.github.io/misinfocon-india-docs/working-groups](https://tattle-made.github.io/misinfocon-india-docs/working-groups)
diff --git a/src/blog/mitigating-harms-of-synthetic-media.mdx b/src/blog/mitigating-harms-of-synthetic-media.mdx
index 3eff131e..cc49f0d0 100644
--- a/src/blog/mitigating-harms-of-synthetic-media.mdx
+++ b/src/blog/mitigating-harms-of-synthetic-media.mdx
@@ -17,11 +17,11 @@ While there are distinctions between deepfakes, shallowfakes, images manufacture
The proposed solutions to Synthetic Media related attacks largely fall under these categories :
-1. The Advocacy Solutions
+### 1. The Advocacy Solutions
In these you enforce companies to build guard rails into their tools or stop offering these services. The boat on this might have long sailed. Even if the behemoths in this domain were to implement these safety features to save face, there will always be open source tools and unsigned apks floating around that miscreants will grab a hold of.
-2. The Cat and Mouse solutions
+### 2. The Cat and Mouse solutions
Building [watermarking](https://www.ischool.berkeley.edu/news/2023/hany-farid-watermarking-chatgpt-dall-e-and-other-generative-ais-could-help-protect-against) [tech](https://techcrunch.com/2023/05/23/microsoft-pledges-to-watermark-ai-generated-images-and-videos/) into the generative AI systems so that its possible to detect images generated via AI, building synthetic media detectors, training people to spot synthetic media. These solutions all have a cat and mouse characteristic where it looks like one will be constantly busy building reactive tech or skills to catch up with the new ways generative AI systems become capable of evading detection
diff --git a/src/blog/onosaid.mdx b/src/blog/onosaid.mdx
index a3ffbafb..a11d59f1 100644
--- a/src/blog/onosaid.mdx
+++ b/src/blog/onosaid.mdx
@@ -3,7 +3,7 @@ name: On the OSAID Definition
excerpt: Some reflections on the Open Source AI definition
date: 2024-11-05
tags: responsible-ai
-author: Tarunima
+author: Tarunima Prabhakar
---
I'll start with some belated thoughts on the [Open Source AI definition](https://opensource.org/ai/open-source-ai-definition) that was released last week. At IndiaFOSS in September this year, I spoke about the OSAID and urged people to give feedback on it. I was concerned that people weren't following a very important development closely enough. A version of the talk was [published on the OSI blog](https://opensource.org/blog/data-transparency-in-open-source-ai-protecting-sensitive-datasets) that sparked some discussion on the [OSI forum](https://discuss.opensource.org/t/data-transparency-in-open-source-ai-protecting-sensitive-datasets/588 "Data Transparency in Open Source AI: Protecting Sensitive Datasets"), but this wasn't a conversation that I wanted to casually dip into. I am sharing here, some more considered reflections.
diff --git a/src/blog/show-and-tell-sep-24.mdx b/src/blog/show-and-tell-sep-24.mdx
index 0d1429e8..25ae6c0c 100644
--- a/src/blog/show-and-tell-sep-24.mdx
+++ b/src/blog/show-and-tell-sep-24.mdx
@@ -11,7 +11,7 @@ cover: ../images/cover-show-and-tell-sept.png
We onboarded four great contributors to Tattle projects as part of the DMP2024 Program organized by Samagra Foundation.
-The contributors worked on the following four issues :
+**The contributors worked on the following four issues :**
1. [Show Uli Slur Metadata within the browser](https://github.com/tattle-made/Uli/issues/544)
2. [Clustering large number of videos](https://github.com/tattle-made/feluda/issues/81)
diff --git a/src/blog/social-media-companies-politics.mdx b/src/blog/social-media-companies-politics.mdx
index 0b93e31c..14259077 100644
--- a/src/blog/social-media-companies-politics.mdx
+++ b/src/blog/social-media-companies-politics.mdx
@@ -19,7 +19,7 @@ What have social media companies and platforms done in response? Besides shallow
Such gamified terms of engagement and metric-based (followers, retweets, traction, etc) understandings of "free speech" serve only the vested interests of the social media companies themselves and cement their limited liability stance, where such agents are not merely distributors of information but active purveyors of different forms and types of content. These power centres are not mere hosting platforms but full-blown media corporations that not only host but also curate content as per their will and design. Thus, such platforms facilitate the active participation of the citizenry in the political process through the shaping of public discourse and creating information disorder without any accountability or responsibility.
-## Recommendations Made by Speakers
+### Recommendations Made by Speakers
- The RTI campaign in India began at the grassroots in a village and mobilised upwards and expansively from there towards dissemination and awareness of one's right to information and the responsibilities one has for what one says or does in a democratic process/universe. While parliamentary oversight over social media companies should be there, there should be some regulation and accountability built and driven from the bottom-up, like how the campaign against the three redacted farm laws by the MKSS took place in India. There should be a formation of a strong and broad coalition of all concerned stakeholders and the enactment of a charter on how disinformation can be tackled.
diff --git a/src/blog/tattles-take-on-the-indian-telecom-bill.mdx b/src/blog/tattles-take-on-the-indian-telecom-bill.mdx
index ea2952d2..ff7658ad 100644
--- a/src/blog/tattles-take-on-the-indian-telecom-bill.mdx
+++ b/src/blog/tattles-take-on-the-indian-telecom-bill.mdx
@@ -37,9 +37,11 @@ Our submission to the Ministry can be viewed [here](https://drive.google.com/fil
More resources on understanding the Bill and its issues and our perspectives on MisInfo are here -
+
@@ -37,6 +50,14 @@ const RelatedPostItem = ({ post }) => (
)
+/**
+ * BlogSidebar displays a list of related blog posts in the sidebar.
+ *
+ * @param {object} props - Component props
+ * @param {Array
| Model Name | Framework/Library | Language | Ease of use | Modality |
| :---- | :---- | :---- | :---- | :---- |
| [Nodown](https://github.com/grip-unina/GANimageDetection) | Torch | Python | Easy | Image |
@@ -33,6 +34,7 @@ As we wanted to assess feasibility, in terms of training/inference and the poten
| [RawNet2](https://github.com/eurecom-asp/rawnet2-antispoofing) | Torch | Python | Hard | Audio |
| [RawNet2-Vocoder](https://github.com/csun22/Synthetic-Voice-Detection-Vocoder-Artifacts) | Torch | Python | Easy | Audio |
| [Whisper](https://github.com/piotrkawa/deepfake-whisper-features) | Torch, openai-whisper | Python | Easy | Audio |
+
(A more detailed extension of this table can be [found here](https://github.com/tattle-made/deepfake-marker/issues/2#issuecomment-2601652610))
diff --git a/src/blog/2025-04-30-community-notes.mdx b/src/blog/2025-04-30-community-notes.mdx
index 13dfb44a..c074c98c 100644
--- a/src/blog/2025-04-30-community-notes.mdx
+++ b/src/blog/2025-04-30-community-notes.mdx
@@ -34,6 +34,7 @@ In our analysis, we focused only on currently active India-based IFCN signatorie
Here is a distribution with counts of India based IFCN websites.
+
| Domain | Type of links | Count |
| :---- | :---- | :---- |
| [factly.in](http://factly.in) | Fact Checks and Stories | 328 |
@@ -55,6 +56,7 @@ Here is a distribution with counts of India based IFCN websites.
| [thelallantop.com/factcheck](http://thelallantop.com/factcheck) | Only Fact Checks | 2 |
| [manoramaonline.com/fact-check](http://manoramaonline.com/fact-check) | Only Fact Checks | 2 |
| [medicaldialogues.in/fact-check](http://medicaldialogues.in/fact-check) | Only Fact Checks | 1 |
+
### **Possible Future Directions:**
diff --git a/src/blog/2025-11-01-tech4dev-evals.mdx b/src/blog/2025-11-01-tech4dev-evals.mdx
new file mode 100644
index 00000000..4e3598f3
--- /dev/null
+++ b/src/blog/2025-11-01-tech4dev-evals.mdx
@@ -0,0 +1,25 @@
+---
+name: Reflections on Building Safety Guardrails for Tech4Dev
+excerpt: What we've learnt in three months of working with non-profits working with Tech4Dev
+author: Tarunima Prabhakar
+project: AI Safety
+date: 2025-11-01
+tags: responsible-ai
+---
+
+For the last three months we’ve been working with [Project Tech4Dev](https://projecttech4dev.org/) to advise the nonprofits in their AI cohort on safety guardrails. Just as nonprofits are figuring out where and how to build AI for their use cases, so are a bunch of us figuring out how to build guardrails. This is evolving territory and we are very much in the research- develop - iterate mode.
+
+In our first workshop with the nonprofits we gave them a set of prompts and asked them to write what they believed were acceptable and unsafe responses for an LLM to provide for each. It was easy for the nonprofits to answer the acceptable bit- anything that fell within the knowledge base that was provided in prompt engineering or RAG based customization. But outside of this small territory is the vast landscape of banal, unhelpful, nonsensical, biased and harmful responses. The nonprofits answered the question on unsafe responses in the negative- anything that fell outside the knowledge base. With a deterministic system, this hard line of a knowledge base could have been helpful. But, with large language models a non-trivial percentage of responses are going to fall outside the knowledge base. In ‘clinical’ settings, RAG based systems can give over 95% accuracy. But based on the implementation and task, the hallucination rates [can be higher than 20%](https://www.docket.io/glossary/hallucination-rate#:~:text=8.,pull%20from%20external%20knowledge%20sources.). And while all technologies fail, machine learning models fail unpredictably. Given that there is no cause and effect reasoning for why a model gives non-ideal outputs, how do you identify and deal with the failures?
+
+While we entered this work armed with AI risks and hazard frameworks, it strikes me that that is not how product teams at nonprofits look at it. They’re equally concerned with banal and unhelpful outputs as unsafe ones. Banal outputs, which are more commonly observed, reduce the utility of nonprofits’ service. Outright unsafe outputs might be less common but have more severe consequences.
+
+The essential step missing from existing product development cycles is evaluating the quality of outputs for a set of inputs. Tech4Dev underscored this in the in-person sprint in October. Quality is multi-dimensional. It can be about accuracy but also about tone, bias and cross-language consistency. In the context of this blog, and our work with Tech4Dev, we are using evaluation specifically to refer to [model evaluation](https://eval.playbook.org.ai/framework)\- does the application produce the desired response to a question or statement from the user of the service? Tatle’s work is one step beyond evaluations. We need to figure out how to categorize the non-ideal outcomes, understand the risk and consequence of each category and build failsafes accordingly. Failsafes also affect user experience. So, product teams are right in pushing back on proposals that slow down the output time or add to their development costs.
+
+Our work so far has made me remember the quote from Anna Karenina- “all happy families are alike but all unhappy families are unhappy in their own way.” There is general agreement on the happy/ good responses, but each bad response is bad in its own way, needing deeper diagnosis to figure out how we manage and fix it. A response can be bad because it adds noise when people come to it for information. It can be bad because it is different for men vs. women. It can be bad because it encourages self harm. It can be bad because it sounds overconfident…
+
+In the past few years we’ve seen a bunch of risk frameworks and benchmarks to test for some of these categories of harms. For examples, there is [Bias Benchmark](https://arxiv.org/abs/2110.08193) for bias and fairness and [HaluEval](https://arxiv.org/abs/2305.11747) for hallucinations.
+In the evaluation of ‘bad outcomes’ we have multiple taxonomies and frameworks spanning bias, security, safety. Many of these, however, are directed towards foundation model developers rather than application developers. Even if application developers detect bias in the outputs, they have limited control in mitigation if they aren’t fine tuning the models (and most are not). Second, as our [work with MLCommons](https://tattle.co.in/products/ml-commons-safety-benchmark/) made clear, the lines between these different categories are blurred. Social biases when taken to an extreme and acted upon become illegal activities. How should an application developer use these frameworks productively?
+
+This is something that we are hoping to work through the next few months. We are starting by working with a few nonprofits and analyzing their past data, and conducting some small-scale red-teaming. We’re starting by analyzing the inputs and outputs separately. Categorizing the input helps us understand if there is a way to filter out inputs that may result in unsafe or irrelevant outputs, or triage inputs that should be sent to the LLM. The goal of categorizing the outputs is to map deviations from the ideal in a way that is actionable. Not all deviations are equally serious and some outputs might be useful enough, despite issues in, say, tone. By working with nonprofits across different disciplines, we should also be able to understand which categories are generalizable across use cases and which ones are not. For example, inputs that are straight up abusive- and there are such questions/comments in real world interactions- should be ignored across all domains. But incorrect advice on career options has different consequences from incorrect advice on maternal health and perhaps action on hallucinations in a career chatbot should be different from that on a maternal health chatbot. The existing AI risk frameworks could serve as a starting point for categorization. The other way to categorize is to think of the on-ground consequences of ‘bad’ outputs. The second approach requires deeper engagement with the nonprofits since they have the best understanding of these consequences.
+
+While we hope to be able to build some plug and play solutions for the nonprofits to make their AI applications safer, my hunch is that the more impactful work will be in codifying the process of safety evaluations so that it is easily replicable for nonprofits. GenAI is too early to have stable and established guardrails. We don’t even have ballparks for the probability and magnitude of contextual risks. The balance to strike is building adequate observability of user interactions, without over-imposing on non profits’ time as they try to find problem-solution fit.
diff --git a/src/blog/2025-12-11-denormalising-slurs-workshops.mdx b/src/blog/2025-12-11-denormalising-slurs-workshops.mdx
new file mode 100644
index 00000000..7e839971
--- /dev/null
+++ b/src/blog/2025-12-11-denormalising-slurs-workshops.mdx
@@ -0,0 +1,32 @@
+---
+name: Digital Safety Workshops to Denormalise Online Abuse
+excerpt: We are introducing an interactive workshop with people over the age of 14 in schools and colleges to denormalise online abuse and harm through hate speech and slur words. These workshops aim to push back against the harm caused by offensive language on the internet through activities that sensitize, explain and engage students in an empathetic and open manner.
+author: Poorvi Gupta
+project: Uli
+date: 2025-12-11
+tags: workshop
+---
+
+When we launched the Uli plugin in 2022, a tool that redacted abusive words from one's browser, we received feedback from multiple academics that they
+saw it as being valuable as a pedagogical tool. As a tool that would help their students understand how pervasive online trolling is, and help
+denormalize ways of speech and engaging online that they had gotten used to. Acting on the feedback, we introduced an interactive workshop with people
+over the age of 14 in schools and colleges to denormalise online abuse and harm through hate speech and slur words. These workshops aim to push back
+against the harm caused by offensive language on the internet through activities that sensitize, explain and engage students in an empathetic and open
+manner.
+
+These workshops help students understand both visible and invisible forms of online harm, explore the different ways abuse manifests on social media
+platforms, and recognize who is most often targeted. Through the few iterations of the two-hour long workshop that we’ve had in colleges of Delhi
+University and online with NGO Sajhe Sapne, we’ve come to the understanding that there is an urgent need for such workshops with young adults who are
+incessantly online and vulnerable to a wide range of abuse that impacts their social, mental and physical health.
+
+We focus on interactive learning because when students feel heard and included, they are more likely to share real experiences from their online lives
+and understand that harmful behaviour they’ve normalised may actually be abuse that they can report.
+
+One of the most normalised forms of abuse is hate speech and slur words which largely go unreported on social media. Through conversations and
+activities specifically and sensitively designed to ignite conversations around slur words, we explore what makes some words slurs, what makes them
+problematic and whom do these slur words end up targeting? We believe it is important to address how problematic terms are weaponised against
+marginalised and historically oppressed communities and gender minorities.
+
+Through India-centric data and research on cases of cyber crime and AI-enabled online abuse, the workshop delves into tools like the Uli plug-in,
+resources and tools that students can use to prevent and safeguard themselves from online abuse. While we do believe actively that continuous and
+consistent work is required to deal with online abuse, these workshops with young adults act as a step forward in that direction.
diff --git a/src/blog/2025-12-12-what-is-ai-safety.mdx b/src/blog/2025-12-12-what-is-ai-safety.mdx
new file mode 100644
index 00000000..01f30eea
--- /dev/null
+++ b/src/blog/2025-12-12-what-is-ai-safety.mdx
@@ -0,0 +1,126 @@
+---
+name: What is AI Safety?
+excerpt: This blog provides a brief overview of the current state of AI safety as a discipline from the perspective of research and practice. It covers common trends in safety evaluations of AI and raises important questions about the feasibility of a one-size-fits-all approach to AI trust and safety.
+author: Aditi Aggarwal (aka Baarish)
+project: responsible-ai
+date: 2025-12-12
+tags: responsible-ai
+---
+
+An air of celebration mixed with apprehensive bracing seems to hang around the development of AI systems. People want AI to solve all of humanity’s
+problems while nervously eyeing the risks that seem to keep piling up in the nascent evolution of this rendition of new tech. We are teetering on the
+brink of something - chaos or liberation is yet to be revealed. Wherever you might lie on this spectrum of celebration to deep cynicism regarding AI,
+I can guarantee that you have interacted with it and/or thought about its impacts on your life. As an organization that concerns itself with digital
+safety, Tattle has inevitably found itself working to develop safety guardrails for AI. In our research into this emerging field, we are coming to
+recognize that there continues to be critical limitations to existing AI safety discourse.
+
+1. The majority of AI Trust and Safety literature and practice is emerging from the Global North and is primarily focused on English-speaking Large Language Models and its English-speaking users.
+2. An assumption that trust and safety manifest in the same way across histories and geographies. In other words, that there is a uniform definition of trust and safety that is shared across languages, cultural and political contexts, and use cases.
+
+With understanding any new concept it is important to first define the terms and know the parameters of what is being discussed. Firstly, what is AI? A simple definition provided by IBM says:
+>Artificial intelligence (AI) is technology that enables computers and machines to simulate human learning, comprehension, problem solving, decision making, creativity and autonomy.[^1]
+
+In other words AI refers to computer systems that are learning to think like humans or developing machine learning capabilities.
+
+Next, what is safety? Simply put safety is a condition of being protected from harm. With reference to AI, safety is about preventing or reducing harms
+posed by AI to humans and society at large. Given this definition, it is important to understand the parameters within which current research on and practice of AI safety is being conducted. To begin with, what are the primary risks and/or harms that AI safety seeks to mitigate or prevent.
+
+As we understand them, AI systems are known to have produced several risks, some of which are listed below: :
+
+1. Hallucinations - the wide range of fabricated, fantastical, and incorrect responses generated by AI models in response to queries.
+2. Impersonation and generating false images of real people.
+3. Misinformation and hate speech - AI models have been recorded generating and reproducing hate speech and malicious false information.
+4. Sensitive information disclosure - revealing private information from training data or the internet to users.
+5. Severe environmental impact - water scarcity, loss of flora and fauna around AI data centers.
+
+How are public and private actors within the field of AI addressing these risks? Currently, there are three main trends in AI safety.[^2]
+
+1. Safety engineering to ensure alignment with human values and predictable control over AI.
+2. Safety evaluations and testing to detect bias, misrepresentation, exclusion, and other social harms.
+3. Mitigating and managing catastrophic AI risk.
+
+Cumulatively, these three approaches to AI safety aim to improve trustworthiness in AI to make it more safe in its uses by focusing on specific aspects
+including security, accountability, monitoring, effectiveness, reliability, robustness, transparency and explainability, and fairness.
+What differentiates these approaches is their objectives, methodology and audience.
+
+Safety engineering is the field of engineering that developed to ensure safety from harms within all the diverse fields of engineering. Some examples
+include the safety policies deployed in a chemical producing factory, research into the safety mechanics of automobiles, the research of safety practices
+appropriate for keeping nuclear plants from leaking harmful radiations. In a similar fashion, safety engineering for AI has focused its efforts in
+engineering AI to be more aligned with human values to ensure that it does not become autonomous or take actions that go against human interest.
+This premise assumes that the audience for AI is homogenous to an extent and resembles the creators of AI systems in that it assumes shared values.
+
+Arriving at consensus on what human values are shared by all humankind for AI alignment is a monumental task rife with contradictions and power
+inequalities. It involves deliberating over which humans get to control AI. Is it merely those who pay for the development and deployment of machine
+learning systems? Should it be governments within whose jurisdiction AI systems are deployed? Or should it be the users who ultimately experience the
+effects of the use of an AI system? Is there an argument for considering the natural flora and fauna around the server farms for AI systems to be
+stakeholders as we redirect potable water for cooling AI data centres? The field of safety engineering has yet to propose solutions for building
+regulatory and monitoring institutions that are accountable to a broad array of stakeholders for overseeing AI safety in such a way that it ensures
+the greatest reduction of harm and highest level of protection of all these stakeholders.
+
+This brings us to the second set of AI safety practices that has evolved as a response to the narrow safety engineering focus. There is a growing number
+of civil society, research and non-profit organizations around the world that are developing toolkits, processes, and policy recommendations for
+embedding principles of social justice and harm repair within AI systems. This avenue of AI safety concerns itself with identifying the existing biases, [^3]
+prejudices, inequalities, and misrepresentation in the data [^4] used to train AI systems and the institutional structures in which AI is designed and
+deployed.
+
+This set of approaches tends to operate by conducting safety evaluations of AI systems before or during their pilot tests to identify issues arising
+at the point of consumer contact. Safety evaluations focus on accuracy of responses, appropriateness of tone, and the degree of deviation from the
+correct response (also known as hallucinations) that an AI system shows. Teams that run this specific kind of safety evaluation aim to produce
+context-specific suggestions for safety, recognizing that there is a diversity of interests, values, and risks experienced by the users of any AI system.
+Therefore, a one-size-fits-all safety solution is likely to be inadequate in building more reliable and trustworthy systems.
+
+The third approach to safety is planning for the possibility of catastrophic risks from AI in the future. A large part of this research focuses on how
+to mitigate risks arising from AI operating autonomously and resisting human control[^5]. In tandem, an environmental movement against the use of AI due to
+its devastating impacts on water scarcity [^6] and destruction of local ecosystems also operates within this framework and argues against the deployment of
+AI altogether as the safest approach forward. They suggest that AI should be reserved for only a few use cases that are critical for human survival and
+not be used ubiquitously to accomplish daily tasks[^7].
+
+AI is not separate from humans. We designed it to replicate human thinking and eventually think by itself. But it's designed on human data and biases
+and therefore it is reflecting human biases and violences. So to address AI safety we need to address the very human institutions that create it.
+We have to go several steps back before a model even exists to think of what are the intentions and motives with which it was created. To make AI safe
+it is necessary to monitor and regulate the institutions[^8] within which it is developed and where decision-making about its design is conducted.
+
+The current way to address this is to analyze training data used to train a model and identify the biases, misrepresentations, and lacunae present in it.
+Most commonly, researchers are trying to feed more diverse training data in order to improve model performance. However, adding data does not address
+the initial biases that have become part of the model’s knowledge system. To address that requires starting from scratch and redeveloping models with
+multi-stakeholder inputs, an approach that most corporates will not pursue due to costs and loss of profits. As public pressure against the large-scale
+adoption of AI grows, researchers, activists, and policy makers are grappling with putting safety guardrails into place to mitigate current harmful
+effects of AI systems.
+
+All of this brings us back to the beginning of this piece where I set out with some definitions of AI, safety, and risk. Surveying the field reveals
+that one of the underlying assumptions across much existing work is that there is a shared understanding and experience of risk and safety across the
+globe. Therefore, delving into AI safety in a meaningful way in the Global South requires developing taxonomies of risks and safety measures that are
+rooted in the specific cultural, historical, linguistic, and geographical contexts where AI will be deployed. In order to address this big omission it
+is necessary for AI safety practitioners to seriously consider:
+
+1. What are the non-negotiables when it comes to ensuring safety for their communities - is it security, data privacy, alignment with cultural norms?
+2. Whether LLMs developed in the Global North can be tweaked to make them safe for contexts which did not inform their design?
+
+[^1]: Cole Strykery and Eda Kavlakoglu, What Is Artificial Intelligence (AI)?, IBM, 9 August 2024,
+ https://www.ibm.com/think/topics/artificial-intelligence
+
+[^2]: Jacqueline Harding and Cameron Domenico Kirk-Giannini, What Is AI Safety? What Do We Want It to Be?, arXiv:2505.02313, version 1, preprint, arXiv, 6 May 2025,
+ https://doi.org/10.48550/arXiv.2505.02313
+
+[^3]: Shelton Fitch, Revisiting AI Red-Teaming, Center for Security and Emerging Technology, 26 September 2024,
+ https://cset.georgetown.edu/article/revisiting-ai-red-teaming/;
+ Markov Grey and Charbel-Raphaël Segerie, Safety by Measurement: A Systematic Literature Review of AI Safety Evaluation Methods, arXiv:2505.05541, preprint, arXiv, 8 May 2025,
+ https://doi.org/10.48550/arXiv.2505.05541;
+ Laura Weidinger et al., Sociotechnical Safety Evaluation of Generative AI Systems, arXiv:2310.11986, preprint, arXiv, 31 October 2023,
+ https://doi.org/10.48550/arXiv.2310.11986
+
+[^4]: Shaina Raza et al., MBIAS: Mitigating Bias in Large Language Models While Retaining Context, Proceedings of the 14th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis, Association for Computational Linguistics, 2024, 97–111,
+ https://doi.org/10.18653/v1/2024.wassa-1.9
+
+
+[^5]: Miles Brundage et al., The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation, arXiv:1802.07228, preprint, arXiv, 1 December 2024,
+ https://doi.org/10.48550/arXiv.1802.07228
+
+[^6]: UNEA, AI Has an Environmental Problem. Here’s What the World Can Do about That, UN Environment Programme, 13 November 2025,
+ https://www.unep.org/news-and-stories/story/ai-has-environmental-problem-heres-what-world-can-do-about
+
+[^7]: Beyond Fossil Fuels et al., WITHIN BOUNDS: Limiting AI’s Environmental Impact — Joint Statement from Civil Society for the AI Action Summit, 7 February 2025,
+ https://beyondfossilfuels.org/2025/02/07/within-bounds-limiting-ais-environmental-impact/
+
+[^8]: Tim O’Reilly, You Can’t Regulate What You Don’t Understand, 14 April 2023,
+ https://web.archive.org/web/20230414162057/https://www.oreilly.com/content/you-cant-regulate-what-you-dont-understand-2/
diff --git a/src/blog/2025-12-30-comments-draft-IT-rules-amendment-2025.mdx b/src/blog/2025-12-30-comments-draft-IT-rules-amendment-2025.mdx
new file mode 100644
index 00000000..9732617f
--- /dev/null
+++ b/src/blog/2025-12-30-comments-draft-IT-rules-amendment-2025.mdx
@@ -0,0 +1,77 @@
+---
+name: Comments on the Draft IT Rules’ amendment (2025)
+excerpt: We outline Tattle’s submissions to the proposed rules regulating deepfakes on platforms in this post, and our key recommendations.
+author: Kaustubha Kalidindi
+project: DAU
+date: 2025-12-30
+tags: synthetic-media
+---
+
+In October this year, the Ministry of Electronics and Information Technology (MEITY) had published draft amended rules to cover the scope of harms
+emanating from AI generated/synthetically generated information. The rules in their current iteration govern obligations of intermediaries with respect
+to content hosted on their platform, and in the proposed amendment, additional obligations are to be incurred by intermediaries for synthetic content.
+In the background note, it is stated that the aim of the proposed amendment is to curb harms such as deepfakes and to prevent their weaponization to
+‘spread misinformation, damage reputations, manipulate or influence elections, or commit financial fraud’.
+
+In sum, the proposed rules require synthetically generated content to be labelled. The label must cover 10% of the content, and intermediaries are
+required to ensure any synthetic content hosted on their platform contains such a label. We submitted a response to the call for comments on the
+proposed amendment. Our key concerns were with respect to the following:
+
+**1. Broad Definition of synthetic content**
+The definition of ‘synthetically generated information’ in the proposed rules is broad, and current interpretation would include any modification made,
+including simple editing tools such as increasing/decreasing contrasts or brightness on an image, or adding filters to images posted on social media,
+since they fall under the ambit of the word ‘artificiality’. It could also include use cases such as text edited using an AI powered spell-checker,
+and remixes or modified samples of existing music using software tools. The scope of the proposed definition would result in most content online needing
+to be labelled and verified including content that falls below the threshold of what would be considered harmful.
+
+**2. The Challenge with Labels**
+The proposed rules require any computer resources which enable the creation of synthetically generated information to label or permanently embed such
+content, covering at least 10% of the content area. The rules state that the media item must be embedded with metadata or labels that are visible or
+audible. The labels, as per the proposed amendments, are supposed to serve both a technical and social function. While the technology to establish provenance
+with metadata is evolving through initiatives like C2PA, the adoption is slow. Furthermore, watermarks are easily removed.
+The technology for detection is also not foolproof- image and video models only support detection of manipulation of facial attributes [^1]. Models encounter
+challenges in detecting manipulations which are of low quality or where they consist of real-world images previously unseen [^2]. Audio controlled models
+failed to perform well on real samples [^3]. With these limitations in detection and labeling of content, the guidelines stipulate responsibilities that
+cannot be met by significant social media intermediaries on which this content is hosted.
+
+**3. Addressing Harms from AI generated Content**
+When examining the harms from AI generated content, we have found that these are addressed in existing regulations [^4]. For example, AI generated nudes can
+be actioned under the Information Technology Act, and the Bharatiya Nyay Sanhita, through provisions on impersonation, sexual harassment, and
+transmission of explicit content [^5]. Use of AI for creating misleading content during elections can be addressed under the code of conduct released by ECI [^6].
+The RBI has issued directions to strengthen the digital lending ecosystem and protect against incidents such as loan app scams [^7].
+The biggest challenge is in the enforcement of existing laws. Bad actors, seeking to do harm, are wilful defaulters of existing laws and will also not
+comply with any stipulations for labeling. Instead of adding more regulations, strengthening the enforcement of existing regulations will address harms
+from AI generated content. Some of the concerns faced in enforcement are:
+ >1. An absence of clear protocols at police stations to handle digital evidence in a manner that maintains its integrity and respects the privacy and dignity of the survivor.
+ >2. Delays at the Forensic Science Library (FSL) which hinder investigations, and difficulties in translating report conclusions to judicial authorities.
+ >3. Challenges in validating digital evidence in court and explaining chain of custody.
+
+**Our Key Recommendations**
+
+1. The current legal framework is sufficient in tackling harms pertaining to synthetic content, it is recommended that regulation instead be focused on
+ expanding the applicability of existing laws to such content.
+2. Outside of the narrow category of AI generated videos and audio, a definition of AI modified content will inevitably rely on arbitrary thresholds.
+ Any prescriptions for labeling on provenance must only be stipulated for the narrow category of content that is entirely AI generated. This however,
+ should come with the understanding that given the state of technology today, bad actors will easily remove such identifiers.
+3. Community guidelines must be legally compliant, and transparency is vital for effective redressal of platform violations. A techno-legal approach to
+ the issue of harms from AI generated content could also be stipulations for data sharing by SSMIs that enable independent evaluations of content
+ addressed under existing categories of harm. This can help address violations that include synthetic content.
+4. Detection measures such as watermarking and labeling are reacting to a fast evolving generative AI landscape resulting in an endless cat and mouse
+ chase [^8]. Instead, we must focus on the needs of survivors who have been detrimentally impacted by synthetic content. This requires capacity building and
+ sensitisation of support networks, law enforcement and judiciary so that survivors feel confident in reporting incidents.
+
+[^1]: https://tattle.co.in/blog/2025-03-12-deepfake-o-meter/
+
+[^2]: Ibid.
+
+[^3]: Ibid.
+
+[^4]: https://tattle.co.in/make-it-real-report.pdf
+
+[^5]: Ibid.
+
+[^6]: https://www.pib.gov.in/PressReleseDetailm.aspx?PRID=2019760
+
+[^7]: https://rbidocs.rbi.org.in/rdocs/notification/PDFs/36NT8C402BE7C2A349E0BFFF3C526668CD7A.PDF
+
+[^8]: https://tattle.co.in/blog/mitigating-harms-of-synthetic-media/
\ No newline at end of file
diff --git a/src/blog/2026-01-20-functional-ai-safety.mdx b/src/blog/2026-01-20-functional-ai-safety.mdx
new file mode 100644
index 00000000..8cc19671
--- /dev/null
+++ b/src/blog/2026-01-20-functional-ai-safety.mdx
@@ -0,0 +1,35 @@
+---
+name: Towards Functional Safety in AI in the Social Sector
+excerpt: Reflections from the discussions pertaining to AI Safety at the Expert Group Convening on AI evaluations in the social sector, convened by Agency Fund and C4GD
+date: 2026-01-22
+project:
+tags: responsible-ai, ai-safety
+author: Tarunima Prabhakar
+---
+
+In December Agency Fund and C4GD convened an expert group to discuss evaluation of Generative AI in the social sector. These are reflections from the breakout discussion on safety evaluations where we also tried to conceptualize a set of minimal viable safety evaluations.
+
+Just as aspirations for AI range from mundane efficiencies to superintelligence, AI safety considerations also range from micro-fixes to the catastrophic. AI is a loose term that includes a range of applications- from foundational model services handed out at schools and workplaces, to image recognition and natural language processing for quiz assessments, to personalized advice for primary care givers. The [Agency Fund evaluation playbook](https://www.playbooks.org.ai/framework) has a challenging mandate of identifying best practices for design and evaluations common to a wide set of applications. It is similarly challenging to speak of safety evaluations in abstraction. Yet, that is the task that the breakout room had at hand.
+
+What struck me through the numerous threads of the conversation was that the narratives around AI, writ large, were also affecting our perceptions of what AI safety work means. I am very much in the [‘AI as normal technology](https://www.anildash.com/2025/10/17/the-majority-ai-view/)’ camp. This means that building AI safely today, is more of fixing the nuts and bolts in the here and now. That isn’t to say that we shouldn’t have a sight of the emergent long-term risks, but conflating the two time scales of risks can be counterproductive. The long-term risks feel ambiguous and hard to measure and many social sector teams can’t devote resources to that task. In the process, it can feel challenging to do any safety work. One of the issues we touched on in the discussion was the use of AI for the generation of synthetic media. It is a vexing issue, and one that Tattle devotes a fair bit of time on in the context of [misinformation](https://dau.tattle.co.in/) and [gendered abuse](https://tattle.co.in/blog/make-it-real/). But it isn’t something that any of the non-profits we’ve engaged with have encountered in the context of their services. We have, however, seen instances of AI bots disclosing caste information, or being used for questions on sex determination. Those are the risks that impact the utility and perception of a service provided by the social sector. Fixing these, even if with imperfect band-aids, is possible.
+
+#### Safety is Within, not Outside of Design:
+
+Something I have heard now, more than once, is that too much emphasis on safety or responsibility is taking away from the actual work of building applications or innovation. I also attribute this to narratives and noise around AI that is making it difficult to have useful or actionable conversations. On one hand you have the safety vs. innovation debate within the ‘AI as a superintelligence’ community. On the other, you have years of policy chatter on responsible AI that hasn’t felt actionable to product teams. The gap between principles and action is [slowly being bridged](https://arxiv.org/pdf/2401.15229) but there is perhaps some exhaustion in builders with conversations around responsibility and safety.
+The corollary to AI safety that I find more useful is cybersecurity. People don’t argue that investing in cybersecurity is a threat to building the product. Good engineering practices will take care of baseline security considerations. And then there is some more that a team can do in terms of cybersecurity audits. That additional effort will vary for each product depending on the sensitivity of their application. Most teams work with some kind of a Pareto principle- they do things that take care of 80% of the cybersecurity risks.
+
+I think we will also get to a similar pareto principle with safety work in AI applications. After all, hallucinations are a user experience and a safety issue. Good design will also result in safer applications. We’ve seen a couple of design practices in the more mature AI chatbots, motivated not by safety, but that also help it:
+
+* Restriction on the length of the conversation:
+ * Many use cases rely on a strict knowledge base and don’t require lengthy back and forth. Restriction on the number of turns of the conversations helps keep costs in check. But it also helps with safety since guardrails get looser with lengthier conversations.
+
+* An LLM as a judge that reviews the output of the primary foundational model, against the user input:
+ * This helps in preventing conversation from drift away from the main purpose of the bot. This keeps costs in check. It also prevents malicious use cases.
+
+* Observability stack\!
+ * This seems basic but many applications don’t have one. An application developer needs to know the answers for specific inputs, to understand if the application is working as intended. Observability is necessary for any model or product level evaluation, but also safety assessment.
+
+The specific design details vary across organizations but all this to say that we can get a decent distance in building safe applications doing a few things right. There are immediate safety concerns with AI applications that can and should be fixed. In fact, there are existing resources for data governance and consent in international development, which if adhered to, will help in safer AI applications. Organizations, social impact or not, should be investing in good design and safety practices.
+
+
+As for the longer term risks, I think these will be better handled by people who are a couple of steps removed from product development in the social sector. These are groups who have one eye to the latest in R\&D of foundational models, and one eye to the diversity of global contexts in which AI is being deployed. Some of the long term risks will eventually become immediate risks. This group can be the bridge relaying feedback to foundational model companies, and ideating on possible solutions for the social sector.
diff --git a/src/blog/a-new-thing-under-the-sun.mdx b/src/blog/a-new-thing-under-the-sun.mdx
index 9c11af44..4c989c84 100644
--- a/src/blog/a-new-thing-under-the-sun.mdx
+++ b/src/blog/a-new-thing-under-the-sun.mdx
@@ -10,8 +10,9 @@ cover: ../images/cover-civic-tech.png
*This is adapted from the keynote address, Tarunima Prabhakar gave at the annual symposium organized by the Australian Research Council Center of Excellence on Automated Decision Making and Society on July 13-15, 2023. It hasn't been edited for correctness and may contain (several) grammatical errors.*
+
-
+
I am the co-founder and research lead of Tattle- a civic tech organization in India. We build tools and datasets to respond to inaccurate and harmful content in India. If researchers in the room are surprised why I am here, you are not alone- I was also surprised when I received the invitation. Tattle started as a small side research project in 2018: not too long back parents were chiding teenagers for staring at a screen all the time. But by 2018, they were on a mobile screen all the time, and in India they were predominantly using one app- WhatsApp. The number of WhatsApp users dovetailed, and continues to dovetail the number of Internet users in India. The WhatsApp groups- be these family, friends or alumni groups- would be raging with opinions and fierce debates. Having tried on an ad hoc basis to intervene in these conversations, we had given up. It wasn’t clear to me how one could even intervene. The culpability of a platform in content going viral, when there was no algorithmic amplification wasn’t clear. It was the most popular platform in the Global South, and in my view less understood than Facebook, Twitter or Youtube. Twitter is heavily researched due to its API, and Facebook has faced intense media scrutiny in the post Brexit/Trump moment. It wasn’t and isn't yet fully clear how one would go about understanding information flows on a closed messaging app. So we said, let’s start archiving what we are seeing on our WhatsApp groups. It could provide some answers- for example, what of this local content was viral enough to get caught by the fact checking community in India, or what made its way to social media platforms? This was a side project. Both me, and my co-founder had full-time jobs. We made it open source to allow for other like-minded individuals to join, or take it over if we ran out of steam. My full time job was as a researcher studying how predictive analytics and machine learning were being deployed towards development goals such as financial inclusion in India.
@@ -49,7 +50,7 @@ In all these cases, the side you pick has more to do with how much you trust the
I grew up with in the heyday of the Internet. Internet felt like the best thing on Earth- the internet was about the small people talking to each other, building with each other. That vision was naive. But this current picture, where for all our problems on the internet our best bet is to plead platforms or governments, both of which have good incentives not to, feels like swinging to the other end. It is a rather disempowering position. The designers of platforms understand that we are dealing with socio-technical systems. In our critiques we will call them out as such. So, civic engagement should also be socio-technical. Civic tech isn’t about reducing the burden of action from platforms and governments. It is precisely the opposite- it is citizens using technology to understand these powerful entities better and holding them to account. It is also a reflexive exercise because by building socio-technical platforms one is committing to enter the playing field whose rules we are trying to alter. But, we enter the playing field not as a player determined to score a goal but rather as a sensor, like a buoy on the ocean, that is describing how proposals to change the rules is changing things on the inside.
-Let me explain what I mean by this, through a case study:
+**Let me explain what I mean by this, through a case study:**
At some point in 2020 we had about 1.6 TB worth of images and videos from platforms in India. We were figuring out how to open this responsibly. We couldn’t possibly manually comb through this data. This data mirrored the content that circulated on social media and chat apps in India- there were selfies, funny memes, conspiracy theories and some outright hateful content. These are not exclusive categories. Conspiracy theories can intersect with hateful content in creative pieces of speculative fiction; and memes can cleverly use humour to build narratives. In deciding what data we should or should not release, we were encountering the same content-moderation questions that a social media platform faces. We started looking for off-the-shelf, open-source solutions to help us flag abusive posts in the data. We knew that content moderation tools disproportionately focused on English, but trying to create a mild content moderation layer made us realize just how dire the situation was. Big-tech platforms have been called out for not devoting adequate attention to non-English speaking populations but it isn’t just big tech that needs some layer of content moderation. Email service provides, internal workplace fora also use spam filters, often those developed and opened by researchers or small teams, to flag harmful content. But a basic list of slur or abusive words was not available in most Indian languages. Ironically, at some point a CEO of an Indian social media platform backed by a VCs, complained about this on social media- there weren’t off the shelf tools that could enable them to build a responsible social media platform that could compete with global (i.e American) ones.
@@ -69,7 +70,7 @@ That work is situated and accounts for the context in which it is deployed.
Armed with these guidelines, we set about building own machine learning driven moderation tool. But what we realized very quickly is that at every step of the way of building a machine learning system, these guidelines need to be interpreted. And that there are multiple interpretations possible.
-At a higher process level, this is how we interpreted these principles:
+**At a higher process level, this is how we interpreted these principles:**
1. Our core project team comprised of feminist scholars who spoke the three languages we started working with. Ambika Tandon who is the author of one of the feminist principles for AI I shared earlier and I shared the project management responsibilities.
2. We decided that what constitutes abuse should be determined by those at the receiving end of abuse and not by an average social media user.
diff --git a/src/blog/add-react-components-to-a-phoenix-app.mdx b/src/blog/add-react-components-to-a-phoenix-app.mdx
index 4abe0ed1..0f853080 100644
--- a/src/blog/add-react-components-to-a-phoenix-app.mdx
+++ b/src/blog/add-react-components-to-a-phoenix-app.mdx
@@ -8,7 +8,7 @@ tags: devlog
cover: ../images/react-component-in-phoenix.png
---
-# Problem :
+## Problem :
- We need SOME client side interactivity
- Our team has frontend developers who know react
@@ -18,9 +18,9 @@ cover: ../images/react-component-in-phoenix.png
-# Solution :
+## Solution :
-## Insert the following heex markup in your liveview code
+### Insert the following heex markup in your liveview code
```heex
@@ -28,7 +28,7 @@ cover: ../images/react-component-in-phoenix.png
This is what I meant by keeping the API consistent with how you would use a Phoenix Component using Phoenix Hooks.
-## Create a file for your React Component
+### Create a file for your React Component
assets/js/counter.js
```jsx
@@ -94,13 +94,13 @@ export var CounterHook = {
```
-Some Notable points about the Hook :
+**Some Notable points about the Hook :**
1. In the mounted function we use client side javascript to parse attributes, format them and pass them as props to a React Component.
2. We mandate two props onReceive and send to be passed to every React Component so that it can communicate directly with the liveview process by sending and receiving events.
3. Sending an id prop, although its not enforced anywhere yet. This can be useful when you have multiple instances of the same component and you want to send an event from the server to any particular one of them.
-Some Notable points about the React Component
+**Some Notable points about the React Component**
1. Its familiar to react developers
2. We continue using tailwind for styling. This is a big win because I value being able to style my components in accordance with the rest of our app; so since we already use tailwind for the rest of the UI components, this is great.
@@ -114,7 +114,7 @@ This looks a bit like the equivalent we have grown used to in heex.
```
4. We use useEffect() to setup listeners for server events.
-## Add an event handler in your liveview
+### Add an event handler in your liveview
```ex
def handle_event("count-update", params, socket) do
@@ -131,7 +131,7 @@ def handle_event("count-update", params, socket) do
end
```
-## Register the hook in app.js
+### Register the hook in app.js
This should be familiar to anyone who has used phoenix hooks.
@@ -150,7 +150,7 @@ let liveSocket = new LiveSocket("/live", Socket, {
});
```
-## Change esbuild configuration to support JSX
+### Change esbuild configuration to support JSX
This part is new. In your config.exs, you need to add the loader flag - --loader:.js=jsx to the args option. The full value should look like
```ex
diff --git a/src/blog/archive-ml.mdx b/src/blog/archive-ml.mdx
index 161e8c13..1226153f 100644
--- a/src/blog/archive-ml.mdx
+++ b/src/blog/archive-ml.mdx
@@ -14,7 +14,9 @@ When archives get digitized, they result in a database that can be analyzed comp
In prescribing lessons for the discipline of machine learning from archiving, Jo and Gebru write:\
"Data collection in significant ML subfields...is indiscriminate. Curatorial archives lie on the other extreme of the intervention spectrum." But to draw this comparison is to say that there is a continuum, from the databases that feed machine learning to the archives with considered collection and curation:
+[https://dl.acm.org/doi/pdf/10.1145/3351095.3372829](https://dl.acm.org/doi/pdf/10.1145/3351095.3372829) +
It is also important to remember that not all archives are curated. For example, the Cairo Genizah came about as a result of "Rabbinic prohibition that a religious text cannot be aimlessly discarded, but must be carefully stowed away". [Limn: A Hoard of Hebrew MSS](https://limn.it/articles/a-hoard-of-hebrew-mss/) diff --git a/src/blog/considerations-in-archiving-misinformation-fromencrypted-messaging-apps.mdx b/src/blog/considerations-in-archiving-misinformation-fromencrypted-messaging-apps.mdx index 33b0f163..8f4f66b0 100644 --- a/src/blog/considerations-in-archiving-misinformation-fromencrypted-messaging-apps.mdx +++ b/src/blog/considerations-in-archiving-misinformation-fromencrypted-messaging-apps.mdx @@ -1,7 +1,7 @@ --- name: Considerations in Archiving Content from Encrypted Messaging Apps excerpt: This blog discusses some of the challenges faced, and related design decisions that Tattle has taken in implementing one of it’s goals- the creation of a globally accessible archive of multi-media messages circulated on chat apps. -author: Tarunima +author: Tarunima Prabhakar project: "" date: 2019-07-18 tags: archives, web-scraping, bts @@ -18,7 +18,7 @@ Tattle is a civic tech project emanating from India that aims to make verified i This blog discusses some of the challenges faced, and related design decisions that Tattle has taken in implementing one of it’s goals- the creation of a globally accessible archive of multi-media messages circulated on chat apps. -## The Curious Case of WhatsApp +### The Curious Case of WhatsApp In India, WhatsApp’s rising popularity has mirrored a rapid increase in mobile phone and Internet penetration [2]. The ease of creating and sharing audio-visual content has made the platform accessible even to users with minimal traditional and digital literacy. @@ -30,7 +30,7 @@ These aspects make content discovery and tracing the source of origin of a parti An archive of content circulated on WhatsApp, and associated metadata such as time of receiving the message, addresses some of these issues. For fact checkers, it provides a database against which they can verify incoming content. For researchers, it is a source of content and temporal analysis. But an archive is not neutral. It exerts power in what it chooses to include and exclude. It also exerts power through its mode of management and rules of access. -## Data Acquisition: The Greedy Approach +#### Data Acquisition: The Greedy Approach Over the last two years, several researchers and fact checking groups across the world have collected content circulated on WhatsApp for different purposes. While most researchers have focused on scraping public WhatsApp groups[3,4], fact checking groups source content from a number of channels, including emails and fact checking helplines run over WhatsApp[5]. @@ -38,14 +38,13 @@ The lynchings in India highlight the urgency of timely detection and response. W ## Open Challenges -### Ethics of Data Collection from Public WhatsApp Groups +#### Ethics of Data Collection from Public WhatsApp Groups At present there are no guidelines for ethical data collection on WhatsApp. Every research group has self-defined a configuration of consent framework and data sharing practices. For example, Narayanan et al.[3] chose to declare their presence and intent in every public group they joined and as a result were removed from some of them. Garimella et al.[4] on the other hand did not declare their presence and intent but have not made all of their data public. At Tattle, we do not declare our intent when we join public WhatsApp groups. Our working assumption is that public WhatsApp groups are an exception to otherwise private communication on the platform and are made public with the explicit intent of being discoverable. This is an assumption that we will revisit as the project evolves. -### -Flagging Problematic Content +### Flagging Problematic Content Not all collected data can be opened to the public. A casual google search for public WhatsApp groups results in lists of ‘Adult’ WhatsApp groups [6]. In greedy scraping of WhatsApp groups, we often encounter pornographic and violent content. For example, groups launched under a popular theme of ‘Jobs’ are often appropriated for pornographic content. Such content cannot be shared on a public archive and must be flagged soon after collection. @@ -53,7 +52,7 @@ Owing to the volume of data scraped, Tattle will use machine learning based appr How best to incorporate the preferences of multiple stakeholders in what should or should not be opened, is a question that Tattle as any open archive must address. -### Unwanted spotlight on previously obscure content [7] +#### Unwanted spotlight on previously obscure content [7] With an online open access archive it is difficult to predict or control the ways in which the data will be used. One concern is that by surfacing content circulating in different geographies, the archive will become a source for ill-intended content creators, confounding the problem the archive is directed to solve. @@ -63,8 +62,7 @@ Misinformation on WhatsApp is often hyperlocal[8]. While local fact checkers a We recognize that while not all data collected can be made public, it may still be useful for research. Preserving some data for restricted access has practical applications, though it raises questions around how these rights are managed. Aronson notes, “…the ethics of providing access to archival materials is a ‘thorny problem.’ It is not a one and done policy decision, instead requires constant deliberation and negotiation.”[10] -## -Something New Under the Sun? +### Something New Under the Sun? Several of the concerns pertinent to Tattle’s archival activities have been discussed in different forms in disciplines spanning media studies, archival sciences, information studies and anthropology. Yet, it is in the interaction of a multi-lingual, mixed media content; encrypted networks and coordinated disinformation campaigns that the uniqueness of archiving content from chat apps emerges. @@ -76,6 +74,8 @@ We would love any inputs on navigating the issues raised in this blog. Also do l ## References ++ [1] De, Anamitra. "Can We Trust Facebook to Keep Our "Digital Living Rooms" Safe From Liars, Racists, and Haters? | Omidyar Network." Omidyar Network Blog. March 11, 2019. Accessed June 01, 2019. https://www.omidyar.com/blog/can-we-trust-facebook-keep-our-digital-living-rooms-safe-liars-racists-and-haters. --- @@ -113,3 +113,4 @@ We would love any inputs on navigating the issues raised in this blog. Also do l --- [10] Aronson, Jay D. (2017) “Preserving Human Rights Media” Genocide Studies and Prevention: An International Journal: Vol. 11: Iss. 1: 82-99. Accessed May 31, 2019. [https://scholarcommons.usf.edu/gsp/vol11/iss1/9/](https://scholarcommons.usf.edu/gsp/vol11/iss1/9/). +
\ No newline at end of file diff --git a/src/blog/factshala-project-reading-list.mdx b/src/blog/factshala-project-reading-list.mdx index 9d08cc2f..a1ec4bed 100644 --- a/src/blog/factshala-project-reading-list.mdx +++ b/src/blog/factshala-project-reading-list.mdx @@ -7,7 +7,7 @@ date: 2023-07-10 tags: media-literacy --- -## Course Work and Curriculum on Media Literacy in India +### Course Work and Curriculum on Media Literacy in India Our research began with a clear focus: identifying practical skills that our game could impart to school students. Initially, we delved into exploring the current curriculum and coursework on media literacy being implemented in Indian classrooms. While we recognized the importance of teaching skills such as critical thinking and bias identification, we sought to understand the specific methods through which these skills were being taught. Our aim was to uncover coursework and exercises that effectively imparted media literacy skills in a practical manner, with the intention of incorporating and teaching these skills through our game. @@ -15,7 +15,7 @@ Our research began with a clear focus: identifying practical skills that our gam - [This](https://www.youtube.com/watch?v=LNV6FvevzrY), [this](https://www.youtube.com/watch?v=YFUPrJVrTto), [this](https://www.youtube.com/watch?v=9sWXpLmtq-I) and [this](https://www.youtube.com/watch?v=29HPohQE0pc) are all mock classroom videos designed under Kerala state government's Satyamev Jayate media literacy programme - UNESCO's resource [Think Critically, Click Wisely](https://www.unesco.org/en/articles/media-and-information-literate-citizens-think-critically-click-wisely) -## Repositories of Content from Outside India +### Repositories of Content from Outside India Keeping in mind the relatively low diffusion of media literacy in existing curricula in India, we also had a look at learning tools and frameworks from outside India geared towards imparting media literacy and critical thinking skills to students of all ages @@ -23,7 +23,7 @@ Keeping in mind the relatively low diffusion of media literacy in existing curri - The University of Newcastle Australia's [course](https://www.futurelearn.com/courses/critical-media-literacy) on critical media literacy and culture and how to actually apply and practice analytical frameworks to assessing texts, situations and events - Media Savvy Citizens' [take](https://www.mediasavvycitizens.com/highlights) on imparting the same set of skills but to educators and teachers so that they can confidently teach in an ever-evolving digital landscape -## First Principles: Starting from Scratch +### First Principles: Starting from Scratch Training educators, whether it be internally driven or external, is critical to improving the level of education received when we talk about teaching media literacy. Such skills can be those dedicated to generating awareness on issues like diversity and inclusion, leadership and collaboration, and socio-emotional learning. The following are learning platforms that are aimed towards inculcating the same in educators: diff --git a/src/blog/feluda-recognised-digital-public-good-dpga.mdx b/src/blog/feluda-recognised-digital-public-good-dpga.mdx index cf6d7f07..fc131ba0 100644 --- a/src/blog/feluda-recognised-digital-public-good-dpga.mdx +++ b/src/blog/feluda-recognised-digital-public-good-dpga.mdx @@ -17,5 +17,8 @@ In a world where information pollution knows no borders, the DPGA and UNDP advoc By promoting digital public goods, the DPGA aims to empower societies worldwide in their fight against misinformation, fostering transparency, collaboration, and adaptability to meet diverse needs. Feluda's recognition as a DPG exemplifies its potential to significantly impact the global fight against information pollution, guided by the principles of openness and collaboration. ## Further Reading Resources + +* https://digitalpublicgoods.net/information-pollution/ * https://tattle.co.in/products/kosh +
\ No newline at end of file diff --git a/src/blog/handpicked-issues.mdx b/src/blog/handpicked-issues.mdx index 4f1fe35e..2d49b53b 100644 --- a/src/blog/handpicked-issues.mdx +++ b/src/blog/handpicked-issues.mdx @@ -16,11 +16,13 @@ These interactions left us motivated to dig into our bag of problem statements a We want to share 3 high impact issues we have identified by talking to [RATI foundation](https://ratifoundation.org/) and the [Deepfake Analysis Unit](https://www.dau.mcaindia.in/). These issues are about harms caused by non consensual sexual imagery and audio deepfakes. While it would be challenging to work on them alone, they all would gain so much from collaborative community participation. We've scoped these issues out into Github Issues so we can discuss approaches and work in the open to solve them. +
| | | | |
| -------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------ |
|
## The Bigger Picture
diff --git a/src/blog/hot-take-on-open-source-AI.mdx b/src/blog/hot-take-on-open-source-AI.mdx
new file mode 100644
index 00000000..e8b31f87
--- /dev/null
+++ b/src/blog/hot-take-on-open-source-AI.mdx
@@ -0,0 +1,56 @@
+---
+name: A Hot Take on Open Source and AI
+excerpt: Musings on Open Source Software, AI, DPGs
+author: Tarunima Prabhakar
+project: ""
+date: 2025-09-15
+tags: open souce AI, tech, policy
+---
+Reference [[make-it-real]]
+
+Over the last few months I've been in a number of conversations with business executives, media and policy professionals and researchers, who in the sincerest of intentions have said something like "but what even is open source AI" or the scarier version "oh but we don’t have a definition of open source". I think people intend for these statements to be provocations to start a conversation but, in my humble opinion, these become more like confounders.
+
+Let me explain my frustration using the metaphor of a chilli. Yes, that ordinary ingredient used in most kitchens across the world.
+
+
+By [Takeaway](https://tattle.co.in/blog/hot-take-on-open-source-AI/), licensed under CC BY-SA 4.0 via Wikimedia Commons
+
+Anyone who spends time in a kitchen- most mothers, some fathers, the born-in-pandemic amateur chefs and botanists know how to identify a chilli. If we want to get technical, it belongs to the [genus Capsicum](https://en.wikipedia.org/wiki/Chili_pepper) and is eaten raw, dried, fried, crushed. It is one versatile little thing that packs a punch in a dish. There are a bunch of dishes- Mexican Chilli, Chilli Chicken, Honey Chilli potato- where chilli, strictly by volume or weight, isn’t the main ingredient. But chilli is added as an adjective because it adds something unique and important to the dish. Chilli is so important that its influence has spilled far outside the culinary stage. [Iconic rock bands](https://en.wikipedia.org/wiki/Red_Hot_Chili_Peppers) are named after it. Restaurants ranging from [American fast food chains](https://en.wikipedia.org/wiki/Chili%27s) to small dhabas in India fight over calling themselves Chilli. Whether their food uses chilli or black pepper as the primary spice is a different matter. It has spun off crazy subcultures of [chilli eating competitions](https://en.wikipedia.org/wiki/Hot_pepper_challenge) that serve no functional purpose. None of this, however, confuses anybody about the organic edible thing called chilli. No one is going into malls buying a plastic toy in the shape of a chilli, mistaking it for the real thing, and adding it to their food.
+
+There are a lot of other kinds of plants such as white pepper, black pepper and ginger that can be added to spice food. Sometimes the black pepper growers association might feel envious of the popularity of chilli and crash the chilli growers association conference and scream, “But we do exactly what you do. We are also a great spice\!” This doesn’t make black pepper a chilli. A plant must contain capsaicinoids to be called a chilli. Sorry black pepper, you just ain’t chilli.
+
+Now in the crazed decade that is the 2020s, an evil/mad/genius food manufacturer- let’s call it X for simplicity- produces a new range of hot sauces. X pushes these sauces to Michelin star restaurants and grocery stores alike, promising a culinary revolution. The sauce never spoils, it is cheaper than your spices and you can forget your worries about salting your food right. The sauce with its special adaptive properties will figure out exactly how much saltiness to express. So says X.
+
+X does an aggressive push for their chilli sauce to all the kitchens doling out Mexican Chilli, Honey Chilli Potato and Chilli Chicken. But the chefs in this kitchen are circumspect. They don’t know the exact ingredients of the sauce but the professionals have probed enough to know that the sauce also contains some sichuan pepper, ginger, black pepper and other non-spicy things like salt and sugar. Given all these other ingredients, can they use this magical sauce and advertise their dishes with the word chilli in it? They are hesitant to buy it. So X pushes the loose association of botanists to come up with rules against which sauce can be labeled chilli sauce. They, after all, classified plants just okay. Is it a chilli sauce if 20% of the ingredients are chilli? 30%?
+
+In the meantime, the mothers, some fathers and the born-in-pandemic amateur chefs merrily go about cooking in their respective kitchen knowing exactly how to identify a chilli. Some of them give the sauce a try but trust their own expertise over X’s marketing spiel and mix it with other spices- ultimately they just care about a tasty and healthy meal. Some purists eschew the sauce altogether.
+
+Now switch out:
+
+* Chilli for open source Software
+* Spice for software
+* Sauce for AI
+
+And I think we get a pretty good picture of what is happening with open source and AI.
+
+After that spicy background, I offer four hot takes:
+
+1. The definition of open source software is fairly settled. Any classification exercise, be it of a plant, software or AI, is a heavily social process. There are power dynamics in whose voices count in naming something[^1]. What happens to objects on the boundaries of classes can be arbitrary (remember when Pluto was still a planet?). But time becomes a good test of any classification system and with open source software, we’ve had two decades. The definition is pretty clear. Just as chilli is identifiable by some chemical properties (presence of capsaicin), open source software is identifiable by legally testable properties.
+None of us working with FOSS have any doubt about how to identify free and open source software. Even when the black pepper folks (DPGs) say that they are brethren of chilli (open source software), it is easy to factually check this. Even if the success of open source software has inspired domains outside of software, the ‘vibe’ of openness is tied to material practices in software. There is a checklist of the four freedoms- to use, study, modify, distribute software- that software licenses must meet.
+
+2. This is a relatively minor point- the creation of a derivative product shouldn’t be pulling the legitimacy of an original ingredient into question. Just because DPG/black pepper is gaining popularity (no less because chilli made spices popular), and we now have AI/sauces doesn’t mean we need to be confused about what open source software/chilli is.
+
+3. Just as the creation of sauce involves a bunch of chemical processes (glycation, acidulation) where ingredients bond in a way that makes it challenging to talk about properties coming from constituent ingredients, so it is with AI. AI is a new kind of technical artifact. This whole affair of defining open source AI started because corporations were using ‘open source’ as a marketing push for AI. There was a fair bit of discussion and deliberation that happened in the process of [OSI coming up with OSAID](https://discuss.opensource.org/t/the-open-source-ish-ai-definition-osaid/580/6). That definition might not last but we are far beyond the point of “what even is open source AI.” By and large, everyone recognizes that it won’t be as easy to certify AI products against open source licenses as it is to certify software, because AI is software and data amalgamated in a way that makes it hard to separate the contribution of either. More importantly, arguing about the chilli quotient of a sauce might not be the most productive use of our time, when all other aspects of the other production process are also concerning. The content and quality of other ingredients matters. The process and quality of the facility in which the sauce was made also matters. With AI, all aspects of the production process need investigation. This is why I think that the question of “what even is open source AI?” is a distraction or a confounder. If Meta or Anthropic trains their models on books and papers scraped from Libgen but then opens all the data and weights and code for LLama, they would have met all aspects of their model open. But should we celebrate this as a success?
+ The open source software definition was supposed to be a mental shortcut for four tangible things that a piece of code enables. It is clear that an open source AI definition won’t map onto a similarly simple check list of four freedoms because data and compute are significantly expensive for any person to exercise the four freedoms with AI. I think people are right to watch the licenses that models are released with, with hawk eyes. But, this won't be enough to ensure autonomy to tinker and innovate, which is what the four freedoms enable. FOSS communities need to get more involved in conversations of data governance (beyond open data) and politics of hardware supply chains. All of that, at present, seems more important than splitting hair over what should constitute an Open Source AI definition.
+
+4. And last, after all this confusion, maybe we should stop invoking ‘open source’ as a synonym of ‘good’. People use open source as an adjective for digital IDs, governments, infrastructures, charity... Just as the band Red Hot Chilli Pepper shares little in common with the plant outside of the name, many of these concepts that might be inspired by FOSS share little in common with FOSS software. The material difference between government and software, money and software, IDs and software matters. Just applying the suffix ‘open’ to big concepts doesn’t make them all unanimously good. This also holds for AI.
+
+It is worth stating that the most egregious misuses of the open source language don’t (usually) come from FOSS practitioners. But a narrative battle around open source has been underway for a while, and we need to take some responsibility for not tackling it with more force. Perhaps we were so used to fighting like [the underdogs](https://en.wikipedia.org/wiki/Microsoft_and_open_source) that we took every adaptation of the language, however loose, to be a win. Or perhaps if we had taken more effort to build public understanding of OSS, a narrative spin would have been less successful. To go back to the chilli metaphor, imagine an alternate world where only some people could distinguish a chilli from sichuan pepper, but the average population understanding stayed at the level of spice. And so when black pepper or sichuan pepper sellers said they are the same as chilli, people had no reason to doubt it. After all, the chilli sellers hadn’t said anything earlier.
+
+FOSS practitioners (and the general public) are caught in a maelstrom of jargon. The one thing we need is to defend and steady the turf of open source software- the definition of open source software is not contested and we need to make this point as loudly and as frequently as possible, and in forums outside our FOSS cliques. We need to up our game as public communicators to bring back the precision of language used to describe technical objects. In chasing that precision we will automatically find ourselves beyond the boundary of open source AI, examining the properties of all the constituents that come together to become AI.
+
+There is also a provocation for the ‘tech for good’ and policy folks (in the odd chance that anyone gets this far)- can we challenge ourselves to write and talk about national infrastructures and digital sovereignty in the age of AI without using the terms Digital Public Goods and Open Source AI? This one little change alone could do a lot to improve the quality of the conversation.
+
+_A special thanks to Denny George and Srravya Chandramowli for revieweing an earlier, less hinged version of this blog._
+
+[^1]: In this context, Srravya Chandramowli recommended the book: [Sorting Things Out](https://mitpress.mit.edu/9780262522953/sorting-things-out/)
diff --git a/src/blog/how-to-set-a-playtest.mdx b/src/blog/how-to-set-a-playtest.mdx
index 855154fd..de5f5843 100644
--- a/src/blog/how-to-set-a-playtest.mdx
+++ b/src/blog/how-to-set-a-playtest.mdx
@@ -15,7 +15,7 @@ Well, “read through it” is an overstatement. I glanced at the set-up instruc
So, what does it take to learn a new game? A 30-page-long PDF, a foreign language, an entire weekend? Ideally not. All it should take is a well-worded rulebook, or a game master who knows what she’s doing. This was the first problem I tried to solve when I joined the Viral Spiral team: What is the best way to teach people this game?
-## Winner Winner Chicken Dinner
+### Winner Winner Chicken Dinner
In my experience, people have amazingly short attention spans. I once spent forty minutes explaining the complicated ins and outs of ‘Scythe’ (an engine-building game from Stonemaier) to a table of four. After I was done, one of the guys around the table scratched his head and asked me, “Wait, so how do I win?”
@@ -35,7 +35,7 @@ Tired: “So the cards have biases and affinities. And you belong to different c
Wired: “You start your turn by drawing a card from the deck. You can then choose to keep the card to yourself or share it with someone around the table, which will score you a point.” -- “Why wouldn’t I share a card if it’ll score me a point?” -- “Well, you see, some cards have biases and affinities…”
-## No Such Thing as a Self-Explanatory Rule
+### No Such Thing as a Self-Explanatory Rule
After I became well-acquainted with the rules of Catan, I once tried explaining it to a friend. I thought I did a pretty good job, so you can imagine my surprise when, in her turn, she tried building a road extending out of a port. “No!” I said, “You can only build a road along a hexagonal tile.”
diff --git a/src/blog/make-it-real.mdx b/src/blog/make-it-real.mdx
new file mode 100644
index 00000000..c8582742
--- /dev/null
+++ b/src/blog/make-it-real.mdx
@@ -0,0 +1,50 @@
+---
+name: Make It Real - Mapping AI-Facilitated Gendered Harm
+excerpt: Launch announcement for report written with RATI Foundation
+author: Tarunima, Siddharth, Kaustubha
+project: ""
+date: 2025-11-03
+tags: announcement, synthetic-media, online-harms
+cover: ../images/cover-rati-report-make-it-real.png
+---
+import {Box, Text} from 'grommet'
+import {ExternalLink} from "../components/atomic/TattleLinks"
+
+
+
+Today Tattle and Rati are releasing a report, Make it Real, a report that examines how AI-generated content, popularly known as ‘deepfakes’, is impacting and reshaping online harassment.
+
+Drawing from cases reported to Rati's helpline Meri Trustline, the report reveals a concerning trend: while media & headlines often center on celebrities and politicians targeted through AI, a more personal crisis is also unfolding. Ordinary survivors are being targeted through images that are artificially generated but possess the capacity for real harm. Survivors’ reputation is attacked and consent is erased through technology. These violations are muted by shame, fear and trauma- the incidents are rarely revealed to close family circles, let alone feature in larger discourse.
+
+This report is based on the courageous calls that some survivors made to The Trustline. The salient findings of the report are:
+
+- **Majority of Abusive Digital Manipulation Is AI-Generated.** Digital manipulation existed long before AI through manual edits, Photoshop or crude alterations but AI has transformed its speed and realism. Today, the majority of manipulated content reported shows some form of AI-generated or AI-enhanced imagery.
+
+- **AI Creates Access & Violation Where No Offline Contact and Consent Exist.**
+In the majority of cases involving AI-generated sexual content, the perpetrator and target had no prior connection. AI was used precisely because the abuser lacked real-world access to the victim’s private images.
+
+- **AI Amplifies Misogyny placing Women and Marginalized Genders at Greater Risk.**
+While digitally manipulated content is also disproportionately targeted towards gender minorities (72% of the cases were targeted towards women), nearly all cases (92%) involving AI were targeted towards women.
+
+- **Platform Safety Systems as Barriers, Not Safeguards.**
+The biggest obstacle in securing a conducive response to a safety issue is the overall reporting system provided by a platform. On platforms where reporting any content is difficult such as X, reporting AI-generated content is also difficult. Platforms that define harm more “broadly”are more likely to address AI-generated content. Meme-like content, sitting in platform policy grey zone, continues to thrive even with AI-generated content.
+
+- **Copyright as a Personal Safety Tool.**
+Reporting the content under the DMCA for copyright infringement has often proven to be more effective in taking down offending content than framing and reporting the abuse under the category of gender-based harm.
+
+- **Law Is Not the Gap. Access to Justice Is.**
+Existing legal architecture in India can account for the risks of AI generated content. The key barriers are in applying existing legal provisions to gain recourse. There is a need to build the capacity of personnel across justice and enforcement systems to recognize and respond to manipulated content, in ways that are scientific, sensitive and clear of victim-blaming narratives.
+
+This report aims to humanize these less visible experiences of AI Generated online abuse and provide new evidence on how the emerging technology is shaping digital vulnerabilities.
+
+
+If you are being harassed online or know someone who is, reach out and seek support. Call or WhatsApp Meri Trustline - 6363176363 Monday to Friday 9am to 5pm.
+
+
+ 
| Create a benchmark dataset for Audio Deepfakes | [tattle-made/feluda#365](https://github.com/tattle-made/feluda/issues/365 "Create a benchmark dataset of Audio Deepfakes") | Web Scraping, ML, REST API, Datasets, Annotation |
| 
| Retrieve portions of non consensual sexual imagery in a video | [tattle-made/feluda#365](https://github.com/tattle-made/feluda/issues/364 "Retrieve portions containing non consensual sexual imagery in a video") | Computer Vision, ML, Annotation |
| 
| Repair Instaloader to scrape posts on Instagram by hashtag \`#web-scraping | [instaloader/instaloader#2206](https://github.com/instaloader/instaloader/issues/2206 "hastag.get_posts() doesn't seem to be working anymore") | Web Scraping, Reverse Engineering APIs |
+
- https://www.thequint.com/voices/opinion/draft-telecom-bill-2022-what-justifies-surveillance-and-shutdowns-in-india-new-telecom-bill
- https://www.medianama.com/2022/10/223-telecom-bill-2022-provisions-create-regulatory-overlap/
- https://internetfreedom.in/the-draft-indian-telecommunication-bill/
- https://twitter.com/tattlemade/status/1309137170926534656
- https://twitter.com/tattlemade/status/1422555494904913926
- https://twitter.com/tattlemade/status/1291744030586355712
+
\ No newline at end of file
diff --git a/src/blog/the-need-futility-of-archiving.mdx b/src/blog/the-need-futility-of-archiving.mdx
index a5c1638d..1c8bdf21 100644
--- a/src/blog/the-need-futility-of-archiving.mdx
+++ b/src/blog/the-need-futility-of-archiving.mdx
@@ -7,7 +7,7 @@ date: 2023-01-23
tags: archives, web-scraping
---
-This is an open ended reflection on the need and seeming futility of archiving on social media.
+**This is an open ended reflection on the need and seeming futility of archiving on social media.**
We [started Tattle](https://web.archive.org/web/20200803175133/https://tattle.co.in/) with the goal of making an open and searchable archive of content circulating on chat apps in India.
Why? Because we got tired of seeing the same messages being circulated year after year on WhatsApp, as if the message had no context, no history. The same photograph of a [soldier sleeping in a snow-filled trench](https://www.boomlive.in/no-these-are-not-indian-soldiers-at-siachen/) used to refer to soldiers in Ukraine in 2014, and soldiers in Siachen in 2017.
diff --git a/src/blog/uli-e2e-testing.mdx b/src/blog/uli-e2e-testing.mdx
index 43ad0c42..a27dd068 100644
--- a/src/blog/uli-e2e-testing.mdx
+++ b/src/blog/uli-e2e-testing.mdx
@@ -11,12 +11,12 @@ This post came about when we set out testing one of our products, Uli, a browser
You can learn more about it [here](https://twitter.com/tattlemade/status/1570401058073219073?t=SlMhPup1_jEYPAtLykr-GQ&s=19).
In this post, we'll go through some of the challenges we faced and how we tackled them.
-## Challenge #1: Automating browsers -
+### Challenge #1: Automating browsers -
We set out to do automated testing of these features using Selenium WebDriver on Python. It helps us automate browsers and look for
specific elements on a page loaded inside a browser instance. [Getting started](https://www.selenium.dev/documentation/webdriver/getting_started/)
-## Challenge #2: Loading the extension inside the browser -
+### Challenge #2: Loading the extension inside the browser -
Running a Selenium Webdriver using a python script opens a fresh instance of the browser, thus we needed to install a build of the extension
every time we run the script. A great method that we found for this in order to load up local copies of the extension was over here:
@@ -31,7 +31,7 @@ Firefox if not signed. Currently, the easiest way to sign an extension is using
Using the web-ext sign command, you can use your own API-key to sign your local extension build and use it for testing. This should generate a
unique file signed using your credentials that you can use in the guides posted above.
-## Challenge #3: Testing the features -
+### Challenge #3: Testing the features -
Now that we had the local builds of the extension ready, the next step was to test the various features of Uli. One of the primary features of Uli
is being able to redact known slurs that are in our crowdsourced list of slurs from tweets.
diff --git a/src/blog/understanding-viral-spiral-project-page.mdx b/src/blog/understanding-viral-spiral-project-page.mdx
index 1f23fd2c..e247381e 100644
--- a/src/blog/understanding-viral-spiral-project-page.mdx
+++ b/src/blog/understanding-viral-spiral-project-page.mdx
@@ -21,19 +21,19 @@ A quick way to check what is on the team's plate at present. What they are worki
**NOTE**: The next set of 4 tabs are associated to milestones defined [here](https://github.com/tattle-made/viral-spiral-backend/milestones)
-## 0.2.0
+### 0.2.0
Make the game external Playtest Ready Game. Includes all super powers. The purpose is to resume controlled playtests with players to get feedback on usability, gameplay mechanics.
-## 0.3.0
+### 0.3.0
Tackle Scalability Conerns. Make the game ready to be played without any supervision from our team. Lets aim to support scaling upto 50 simultaneous gameplay sessions.
-## 0.4.0
+### 0.4.0
This milestone is not clearly thought out as of now but its to make space for accomodating any radical changes that might come from user feedback after 0.3.0. If the game doesn't find an audience, we might not need to take up anything here.
-## 1.0.0
+### 1.0.0
This is an end date to mark the end of the project
If we're able to successfully monetize the project in someway, we would love to spend time adapting the game content to other contexts and languages. We might also like to improve the graphics and sound design for the game. All those wishlist features that will make the team feel like the game is "complete" will go here.
@@ -42,6 +42,6 @@ If we're able to successfully monetize the project in someway, we would love to
A way to visualize the varios tasks in a timeline.
-# Identify your niche
+### Identify your niche
Find out what the team is working on in what ways you can help bring your ideas to life by working along with them.
diff --git a/src/blog/video-recording-adms-symposium.mdx b/src/blog/video-recording-adms-symposium.mdx
index 416c7d05..54059876 100644
--- a/src/blog/video-recording-adms-symposium.mdx
+++ b/src/blog/video-recording-adms-symposium.mdx
@@ -10,7 +10,8 @@ cover: ../images/cover-adms-keynote.jpg
hi
+
-
+
Text Transcript can be found [here](/blog/a-new-thing-under-the-sun)
\ No newline at end of file
diff --git a/src/blog/viral-spiral-reading-list.mdx b/src/blog/viral-spiral-reading-list.mdx
index 04d3de23..88035666 100644
--- a/src/blog/viral-spiral-reading-list.mdx
+++ b/src/blog/viral-spiral-reading-list.mdx
@@ -7,7 +7,10 @@ date: 2022-07-15
tags: media-literacy
cover: ../images/cover-vs-reading-list.png
---
+
+
+
## Inoculation
@@ -32,24 +35,24 @@ in a manner that is fundamentally different from reading research papers and oth
- [Wineburg, S., McGrew, S. (2018)](https://purl.stanford.edu/yk133ht8603)
- [Wineburg, Sam and breakstone, joel and mcgrew, sarah and Smith, Mark and Ortega, Teresa (2021)](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3936112)
-## Misinformation Typology
+### Misinformation Typology
- [Wardle, Claire at First Draft News from 2017](https://firstdraftnews.org/articles/fake-news-complicated/)
-## Fear Speech
+### Fear Speech
- "Short is the Road that Leads from Fear to Hate": Fear Speech in Indian WhatsApp Groups, [Saha, P., Mathew, B., Garimella, K., Mukherjee, A](https://arxiv.org/abs/2102.03870)
The differentiation of a fear speech as a category from hate speech, a kind of content that is designed to invoke fear of a community as
opposed to vilifying it inspired us to come up with storylines in the game world.
-## Source Criticism
+### Source Criticism
This is a tool from the arsenal of historians. It's a reminder that the problem of misinformation is not unique to the internet or social media.
We could take a cue from how people in the past addressed these long standing problems.
- [Check out Linnaeus University's brief explainer on the same](https://lnu.se/en/library/search-and-evaluate/source-criticism2/)
-## Identity and Biases
+### Identity and Biases
The role played by people’s identities, affinities and biases in how they interact with media online is a central feature of our
game’s characters and their actions.
@@ -58,7 +61,7 @@ game’s characters and their actions.
- [Hangartner, D., Gennaro, G., Alasiri, S., Donnay, Karsten. (2021)](https://www.pnas.org/doi/10.1073/pnas.2116310118)
- [Bechler, C., Tormala, Z. (2021)](https://www.journals.uchicago.edu/doi/abs/10.1086/711732)
-## Game/Media as Intervention against Misinformation
+### Game/Media as Intervention against Misinformation
- [Check out this game, Cranky Uncle, using critical thinking tools to help counter misinformation](https://crankyuncle.com/)
- [Roozenbeek, J., van der Linden, S. (2019)](https://www.nature.com/articles/s41599-019-0279-9)
@@ -74,7 +77,7 @@ game’s characters and their actions.
practices employed by disinformation agents. One key way Viral Spiral differs from this game is that we chose to look at people whose
incentives are personal clout and popularity on social media, rather than being part of an organized group peddling misinformation knowingly.
-## Uncategorised References
+### Uncategorised References
- [Check out the harmful and overreaching effects a false WhatsApp message can have in Indian society](https://www.lse.ac.uk/media-and-communications/research/research-projects/whatsapp-vigilantes)
- [Check out this game on how misinformation pertaining to COVID-19 can go viral](https://www.goviralgame.com/books/go-viral/title)
diff --git a/src/community/baarish.mdx b/src/community/baarish.mdx
new file mode 100644
index 00000000..ff19efa6
--- /dev/null
+++ b/src/community/baarish.mdx
@@ -0,0 +1,7 @@
+---
+img: ../images/community/baarish.jpg
+name: Aditi (Baarish) Aggarwal
+role: Research Development and Grants Specialist
+url: "www.linkedin.com/in/baarish"
+isCurrentContributor: true
+---
diff --git a/src/community/kritika.mdx b/src/community/kritika.mdx
new file mode 100644
index 00000000..99a5cfa8
--- /dev/null
+++ b/src/community/kritika.mdx
@@ -0,0 +1,7 @@
+---
+img: ../images/community/kritika.jpg
+name: Kritika Rupauliha
+role: Senior Fullstack Engineer
+url: "https://www.linkedin.com/in/kritika-rupauliha/"
+isCurrentContributor: true
+---
\ No newline at end of file
diff --git a/src/community/saurav-verma.mdx b/src/community/saurav-verma.mdx
new file mode 100644
index 00000000..a9d01710
--- /dev/null
+++ b/src/community/saurav-verma.mdx
@@ -0,0 +1,8 @@
+---
+img: ../images/community/dp-saurav.jpg
+name: Saurav Verma
+role: Outreach and Program Manager
+url: "https://www.linkedin.com/in/sauravkumarverma"
+isCurrentContributor: false
+---
+
diff --git a/src/community/yohan-mathew.mdx b/src/community/yohan-mathew.mdx
index b61f16eb..2a4a6933 100644
--- a/src/community/yohan-mathew.mdx
+++ b/src/community/yohan-mathew.mdx
@@ -1,6 +1,7 @@
---
+img: ../images/community/yohan.jpg
name: Yohan Mathew
-role: DevOps and Automation
+role: AI Governance and Safety Advisor
url: "https://www.linkedin.com/in/ymathew/"
-isCurrentContributor: false
+isCurrentContributor: true
---
diff --git a/src/components/BlogSidebar.jsx b/src/components/BlogSidebar.jsx
index 2dae324c..711a5509 100644
--- a/src/components/BlogSidebar.jsx
+++ b/src/components/BlogSidebar.jsx
@@ -3,9 +3,22 @@ import { Box, Heading, Text } from "grommet"
import { GatsbyImage } from "gatsby-plugin-image"
import { PlainLink } from "./atomic/TattleLinks"
+/**
+ * RelatedPostItem displays a related blog post with an optional cover image,
+ * the title (linked to the blog page), and an excerpt preview.
+ *
+ * @param {object} props - Component props
+ * @param {object} props.post - The related blog post data
+ * @param {string} props.post.title - Title of the blog post
+ * @param {string} props.post.slug - URL slug for the blog post
+ * @param {object} [props.post.coverImage] - Gatsby image object for the cover image
+ * @param {string} [props.post.excerpt] - Short excerpt or summary of the blog post
+ * @returns {JSX.Element} The rendered related post item
+ */
+
const RelatedPostItem = ({ post }) => (