Q3 2024 at Tattle. It is All About the People.

One would have expected things to become quieter post the Indian national elections but the third quarter of 2024 was its own kind of busy. The theme for 2024 Q3 became expanding and building community around our projects.

Project Updates

Deepfakes Analysis Unit Dashboard:

• Till June of this year, we were in extreme vigilance and firefighting mode, building and maintaining a first of its kind collaborative tipline for AI generated content. With the elections behind us, in the third quarter we focused on reviewing usage of the tipline to see the kind of queries coming in, to prioritize solutions. By reviewing user chat history, we noticed that many curious users who were engaging with the tipline were struggling to understand the project's intent and how to use it. This led to us conceptualizing a first of its kind User Research of Whatsapp tipline users.

  We also addressed a pain point that the DAU secretariat shared of detecting duplicate or similar queries so that they could be responded to automatically. Both these projects were worked upon with a community as part of Code4GovTech's Digital Mentorship Program 2024. Details of the project are listed below.

• We conducted the analysis for DAU's first quarter report. The report provides some interesting signals on the prevalence of AI generated content on social media.

Uli:

• Uli was officially recognized as a Digital Public Good by the Digital Public Goods Alliance!

• We are partnering with Access Polity on their non-partisan program to support campaigners keen on centering gender issues in political narratives. We created a module for the participants on responsible use of tech and AI, set against the current receiption of women politicians online.

• Safety Benchmark Datasets with MLCommons:

  We were one of the pilot projects selected by MLCommons to contribute to their v1 safety benchmark dataset. For this pilot we're testing the expert led participatory methodology that we piloted with Uli, to create more contextually relevant safety datasets for large language models.

Tattle in the Community

• Code4GovTech Program:

  Through this quarter we were mentoring four volunteers- three engineering students, and one UX designer- who worked on improving three of our projects - Feluda, DAU and Uli. Read on to learn more about their contributions:

  • Clustering Large Amount of Video and Audio -

    When studying large volume of data spreading on social media, it is sometimes useful to ask broad exporatory questions like "Show me videos of people doing a dance trend" or "Show me videos of a political rally" or even "Show me all good morning messages". Snehil and Chaithanya evaluated different clustering algorithms, vision language models (VLMs) and created a custom video dataset focused on Indian context, for future experiments. The final workflow was merged into Feluda, making this available to everyone in the community and specifically the DAU project and Uli.

  • Show Slur Metadata in the Uli Plugin -

    Uli users have shown some excitement around using the plugin for pedagogical purposes that is to inform people about why a word could be hurtful. We had access to this information from our expert contributions. Hardik's contribution makes it possible for Uli plugin users to see this information in their browser!

  • As mentioned earlier, we perceived that the intent of the DAU tipline wasn't always clear to users. Tanushree interviewed fifteen people that included fact checkers, WhatsApp tipline users and non tipline users to understand users' expectation of tiplines, and how the DAU tipline can be better utilized for reporting content.

• We had a blast at India FOSS 3.0 in a corner with other incredible open source for social impact organizations. The zines kept running out but the participation energy did not wane. We also spoke on a panel on open source in AI, and gave a lightning talk on OSI's definition for open source AI. A summary of the presentation can be found on the OSI blog.

• We spoke at TrustCon 2024 about participatory approaches to building representative datasets for moderation. We also participated in a roundtable on Language Support, Tooling, and Content Moderation.

• Uli was featured by Github on their Github Open Source Friday segment. We discussed how Uli began as a collective response to OGBV, demonstrated how the plugin works and what's in store for the project.

• We participated in DFL and POV's workshop on building Gender Equitable Futures, called Feminist Futures Now, where we discussed strategies to advance digital rights and safety for marginalised groups amongst activists and organisations working with women, trans and queer communities, sex workers, and persons with disabilities at the intersections of tech and society in India.

• We participated in Aapti Institute's convening of the AI x Safety Community of Practice (CoP) which was launched earlier this year, and discussed how AI can be leveraged to promote cybersecurity and digital integrity. We later also attended their event on Shaping AI Futures: Blueprints for the Responsible AI Ecosystem in India, which focused on the evolving landscape around Responsible AI in the country.

The unintended benefit of sending this newsletter two months into the last quarter of the year is that we can tell you with certainty that all the community engagement with Tattle yields concrete product outcomes. We'll have more to say in our last newsletter of the year!

Till then, wishing you the best!