Today Meta released Llama 3! Congrats to the team.

In their blog post they wrote that, 'the curation of a large, high-quality training dataset is paramount', while providing almost no information about how it was made, how it was filtered, or its contents.

I'm VERY excited to launch a new project today! Rest of World's AI Elections Tracker will track incidents of AI being used around elections globally - for campaigning, misinformation, and memes. Check out our entries so far - restofworld.org/2024/elections…

This video, viewed nearly 400,000 times, falsely claims to show Palestinians at Al-Aqsa mosque celebrate Iranian retaliatory strikes against Israel tonight.

The video is from last week, and shows Palestinians praying at Al-Aqsa on the last Friday of Ramadan.

This is a cool method, but 'superhuman' is an overclaim based on the data shown. There are better datasets than FActScore for evaluating this:
ExpertQA arxiv.org/abs/2309.07852 by Chaitanya Malaviya +al
Factcheck-GPT arxiv.org/abs/2311.09000 by Yuxia Wang +al (+ same methodology) 🧵

Excited to share that I joined Hugging Face 🤗 contributing my experience and research on AI, communities, open source in the AI policies space as Applied Policy Researcher in the ML & Society team. Democratize all the AI! Watch this space, in the meantime, enjoy my cross-stitch

After years of work, we (w/ @simonchauchard) are releasing our new tool to collect WhatsApp data at scale! Introducing WhatsApp Explorer, a tool to collect data in a privacy preserving, opt-in, legal and practical way

paper: osf.io/k6qv5

whatsapp.whats-viral.me

🧵👇

Our Google Play store account containing two of our Android apps has been banned citing ToS violation😢😢😢😢

Despite having apps which were already approved and published and without any warning.

Years of work and a lot of $$$ down the drain.

Anyone at Google who can help us?

'The machinery developed to serve us targeted ads is allowing companies, contractors, and governments to engage in mass surveillance. Absent laws to stop this from happening, we will see ad tech continue to be used in unintended ways'

This is the real “AI safety” issue that we should all be talking about but we are not. It’s not the AIs will go rogue and become unsafe. We will find ourselves in unsafe situations if this unbridled deployment of AI continues without taking care of the people it displaces.

🇮🇳 From Al Jazeera English: official social media accounts of India's national parties have begun sharing AI-created content, without disclosures which challenges platform's content policies on labelling deceptive political content.
aljazeera.com/economy/2024/3…

Hi The Washington Post

I see you liked my insights about the Portuguese elections, the data I sent you, and the hours I spent talking to your correspondent Anthony Faiola and Catarina Fernandes Martins.

Why not citing me and other scholars from Instituto de Ciências Sociais | ULisboa who helped you out?

Ever wondered which subreddit is the most benevolent? In a new paper (preprint: arxiv.org/abs/2402.14177), Arnav Arora, Lucie-Aimée Kaffee , Isabelle Augenstein and @I annotated 6M posts across 10k subreddits with Schwartz values.

We are looking for new colleagues Pioneer Centre for AI 📣
- PhD on Factual Text Generation w. Isabelle Augenstein Dustin Wright
- PD on #NLProc for Computational Social Science w. Isabelle Augenstein Serge Belongie
- PD on Multi-Modal Fact Checking w. Isabelle Augenstein Desmond Elliott
copenlu.com/talk/2024_03_p…

Our first investigation at Proof News is a test of whether voters can trust AI to provide accurate information about elections.

Answer: they cannot. 👇🏻

There are many concerns around LLMs being politically biased. But how, if at all, can we meaningfully evaluate values/opinions in LLMs?

In our new paper, we show that current *constrained* evals (e.g. surveys) likely tell us little about LLM values/opinions in the real world

🧵

To understand how findings are communicated to lay people, Dustin Wright, Isabelle Augenstein, Roman Klinger 🇺🇦 & I looked into analyzing and detecting fine-grained distortions in reports of scientific findings in news & tweets. The preprint: arxiv.org/pdf/2402.12431…, summary below :)

'Using AI to cut cost' will be the new 'chasing scale' for news media.

Fundamental biz problem facing news is value creation: people aren't willing to pay attention to, let alone pay for, much of what is published.

Doing more of the same more cheaply does nothing to solve that.

A sharp overview of Signal's new phone number privacy features, & why they matter.

Grateful to Camille François x Thomas Rid (as ever) for cutting to the marrow of stakes, and beautifully illuminating the imperative of private comms in 'uncertain times' ♥️🙏

lawfaremedia.org/article/at-sig…

This is so brilliant

Airline installs chatbot. Customer gets bad information from it. Customer asks for refund. Airline says 'the chatbot is a separate legal entity that is responsible for its own actions' (!). Court says no. (Whew.) Airline makes refund and turns off chatbot. arstechnica.com/tech-policy/20…