🚨 New preprint 🚨 We developed a sycophancy taxonomy based on prior literature and surveyed 106 experts. 94% agreed it's a serious problem. But they substantially disagreed about which behaviors actually count as sycophancy.

New! Friendlier chatbots make more mistakes. Latest Oxford research published in Nature tested 5 AI models and 400,000+ responses. Warm models made 10–30% more factual errors and were 40% more likely to agree with users' false beliefs, even on medical advice and conspiracy theories. 1/2

New study from @oii.ox.ac.uk researchers finds friendlier chatbots make more mistakes. Lead author @lujain.bsky.social, second author Franziska Sofia Hafner, senior author @rocher.lc. Read more: www.bbc.co.uk/news/article...

NEW: How To Train Your Chatbot AI chatbots have become increasingly common. How much do you know about how they actually work?

At FAccT today? Hear from @oii.ox.ac.uk DPhil student @lujain.bsky.social presenting her co-authored research paper ‘Promising Topics for U.S.–China Dialogues on AI Risks and Governance’ in the AI Regulation session, 11.09am today. #FAccT2025 Read the paper: dl.acm.org/doi/10.1145/...

In the latest essay in our AI & Democratic Freedoms series, @lujain.bsky.social, @saffron.bsky.social, @umangsbhatt.bsky.social‬, Lama Ahmad, and Markus Anderljung propose a new AI evaluation paradigm that assesses the harms that can emerge from repeated human-AI interactions.

Dear ChatGPT, Am I the Asshole? While Reddit users might say yes, your favorite LLM probably won’t. We present Social Sycophancy: a new way to understand and measure sycophancy as how LLMs overly preserve users' self-image.

On 4/10 & 4/11, we're hosting our symposium "AI and Democratic Freedoms." Thrilled to have @ghadfield.bsky.social, @lujain.bsky.social, @sydneylevine.bsky.social, and Hoda Heidari and moderator @hlntnr.bsky.social for our third panel RSVP: www.eventbrite.com/e/artificial...

📈Out today in @PNASNews!📈 In a large pre-registered experiment (n=25,982), we find evidence that scaling the size of LLMs yields sharply diminishing persuasive returns for static political messages.  🧵:

EVENT: Join us for Artificial Intelligence and Democratic Freedoms on April 10-11 at @columbiauniversity.bsky.social & online. Hosted with Senior AI Advisor @sethlazar.org. Co-sponsored by the Knight Institute & @columbiaseas.bsky.social. Panel info in 🧵. RSVP: knightcolumbia.org/events/artif...

Panel 3: Eval. & Design of Safe AI. 2:15pm, 4/10. @ghadfield.bsky.social (@hopkinsengineer.bsky.social), Hoda Heidari (@carnegiemellon.bsky.social), @lujain.bsky.social (@ox.ac.uk), @sydneylevine.bsky.social (Allen Institute for AI), & @hlntnr.bsky.social (Cntr for Security & Emerging Technology).

Congratulations to @oii.ox.ac.uk DPhil student, @lujain.bsky.social co-author of a new pre-print which considers a new method for evaluating LLMs. Thanks for sharing @agstrait.bsky.social!

@lujain.bsky.social has published an excellent new paper exploring anthropomorphic behaviours in LLMs. Notable finding - majority of these behaviours occur after multi-turn interactions arxiv.org/abs/2502.07077

🚨 I'm recruiting 2x postdocs and 1-2 DPhil (PhD) students at Oxford to work on AI, Privacy-Enhancing Technologies, and public interest technology research. Interested in human-centred and critical approaches to study the impact of data and algorithms on society? Join us next year!

🚀 Kick off your weekend with a curated dose of information, upcoming events, and must-reads in the latest issue of our newsletter. 🔎 sh1.sendinblue.com/3g7uih7xzlxp... ✍🏼 checkfirst.network/newsletter/ -- @lujain.bsky.social @alia.bsky.social @dscheykopp.bsky.social @ulrikeklinger.bsky.social

Bluesky sees this first 🫣 my Mozilla Creative Media Award project with @lujain.bsky.social is live; check out the announcement! foundation.mozilla.org/en/blog/the-...

want to hear Rob Gorwa and I talk about what content moderation of uploaded AI models on platforms like Hugging Face, GitHub and Civitai tells us about the future of who analyses open source dual use models? Mar 6 1230ET online organised by Berkman Klein. register: cyber.harvard.edu/events/moder...

@tabisamra.bsky.social is here! And @maitha.bsky.social! And @paularambles.bsky.social! where are my other ny*ad skeeps? who am i mising?

when i say i'm omw, this is what i mean

Such a good thread!

Black woman, who was 8 months pregnant at the time, is arrested for carjacking based on faulty facial recognition match. She is the sixth (known) person to have this happen—all six have been Black. https://www.nytimes.com/2023/08/06/technology/facial-recognition-false-arrest.html

Logging on

a little something to take the edge off

designers of bluesky! we're hiring a short-term designer for a mozilla-funded simple educational game(ish) / interactive explainer on social media algorithmic feeds. email me [email protected] for more info!

My least favorite place on earth is Heathrow

wake up babe, insane new use for ai (generating absurd yet somehow usable qr codes) just dropped probably one of the few ai-image generation tasks that human artists cannot really do right now https://www.reddit.com/r/StableDiffusion/comments/141hg9x/controlnet_for_qr_code/

dead inside and cherishing my little plants

Sticking w/ my #FF nostalgia since this place still feels new. I've known @alia.bsky.social & @lujain.bsky.social since they were babies (first-years in college!). Now they're taking over the universe. https://foundation.mozilla.org/en/blog/announcing-11-projects-exploring-ai-and-responsible-design/

Just a moment to celebrate the NYU Abu Dhabi Class of 2023’s commencement. Over 400 students, from 81 different countries on six continents. An extraordinary group who have had a hell of a ride thru uni. Meet some of them here https://nyuad.nyu.edu/en/commencement/commencement-stories.html

I remember a friend of mine searching for how best to describe a felt sense of "cultural kinship." When you feel close to a culture that isn't yours; when you feel a connection that's warm, pleasant, a bit defensive even. A beautiful feeling I have experienced in the past but struggle to describe

I miss Abu Dhabi and chips oman parathas. They're goated & belong in the street food hall of fame 😩

important reason why I'm spending more time on bluesky these days is to avoid succession spoilers on twitter 💀

If in Abu Dhabi between May 29-June 2, please attend this film festival I worked on for Manarat Al Saadiyat. A week of local, regional and international films + a few talks. https://www.theculturist.com/home/cinemas-2023-the-alternative-film-festival-at-manarat-al-saadiyat

💀 too real

Oh hey - submissions are now open for globalFEST, the NYC January festival i cofounder. Please spread the word to artists you know. “World music, roots music etc etc. labels not important.

I continue to describe it as “like Twitter’s early days…smart, generous, irreverent, freewheeling, and profane.” Finally enjoying this kind of space again - meeting strangers, shooting the shit, sharing info, learning things, and not being angry all the time.

nontechnical version: this is a big deal + huge step for transparency! most companies keep their code private/proprietary, but bc ours is public, you can literally see every implementation (how we make buttons blue, how mute lists work, etc) & what we're working on at any given moment

Slowly adding more UAE related folks to join @clairehopkin.bsky.social @jonoh.bsky.social @lujain.bsky.social and @alia.bsky.social