🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

awesome work from the brilliant, tenacious & impeccably organized Jannik Kossen ! works great for contrastive learning, but generally this is a performant + flexible way to distill from teachers when you don't want the student to be too forced to exactly match. go chat to Jannik!

I'm hiring! I'm building 4 research groups under me at AISI (formerly the UK's Taskforce on Frontier AI) to work on foundational AI safety research. [1/5] gov.uk/government/pub…

Want to catch up with the rapid progress in ML for functional protein design? Not sure where to start? Check out our review in Nature Biotech! #ProteinDesign #NatureBiotechnology #Cover

The current default recipe for Bayesian active learning doesn’t really work beyond MNIST scale. We suggest why that is and identify a simple fix. arxiv.org/abs/2404.17249 AISTATS Conference with @adamefoster Tom Rainforth 1/5

I have an opening for a 2.5-year postdoc position in the RainML lab as part of my ERC grant on probabilistic machine learning and intelligent data acquisition. Application deadline 10th July 2024. See here for details and to apply: tinyurl.com/rainmlpostdoc

Great piece in TIME on hallucinations and semantic entropy from Billy Perrigo Precise but accessible. Was a pleasure to speak with him. So many insightful questions.

Our work on detecting hallucinations in LLMs just got published in nature! Check it out :)

Excellent piece by karin verspoor (professor) in @nature News and Views discussing our recent paper on detecting hallucinations with semantic entropy nature.com/articles/d4158…

I was invited by Nature News & Views to comment on a paper recently published in nature by Sebastian Farquhar Jannik Kossen Lorenz Kuhn Yarin on "Detecting hallucinations in large language models using semantic entropy". My piece on LLMs "Fighting fire with fire" is here: rdcu.be/dLkVv

Major study out now in Nature by Prof Yarin Gal Yarin, Dr Sebastian Farquhar Sebastian Farquhar, Jannik Kossen Jannik Kossen & Lorenz Kuhn Lorenz Kuhn advances the reliability of AI with novel approach to improving the detection of 'hallucinating’ LLMs. Read more: cs.ox.ac.uk/news/2345-full…

Semantic Entropy Probes: Robust and Cheap Hallucination Detection in LLMs abs: arxiv.org/abs/2406.15927 Semantic entropy: repeated LLM generation at non-zero temperature is usually consistent for accurate generation, inconsistent for inaccurate generation. This consistency can

Yoav Artzi Sewon Min I think this paper by Jannik Kossen might be what you are looking for: openreview.net/forum?id=YPIA7…. We also looked at this in the many-shot setting on a sentiment analysis task (that models do look at labels indeed and can even override pre-training biases): arxiv.org/abs/2404.11018

Life update! I've joined FAIR @meta as an AI Research Scientist to work on code generation with LLMs in Gabriel Synnaeve's team ✨ Thanks to everyone who supported me along the way 🙏 I'm super excited for what's to come!

hello ! we will be presenting Estimating the Hallucination Rate of Generative AI at NeurIPS this friday come if you'd like to chat about predicting and understanding in-context hallucinations and epistemic uncertainty poster 2703 - east - 4:30pm arxiv.org/abs/2406.07457

The aleatoric-epistemic view on uncertainty doesn't serve ML researchers' needs and should be replaced. Come to the talk and poster tomorrow (Sat 14 Dec) at the #NeurIPS2024 workshop on Bayesian decisions (gp-seminar-series.github.io/neurips-2024). openreview.net/forum?id=WIjgb…