Forthcoming book on the human side of AI (and the hype that keeps them hidden): Spring 2025
LLM Reads with the Human Feedback Foundation
In our quest to understand the current wave of activity in Natural Language Processing (NLP), and in particular Large Language Models (LLM), we are happy to launch our second season of talks.
We invite the first authors from some seminal LLM papers to talk through their paper, explore any common misconceptions they’ve encountered, and lead a discussion on how things have changed since the paper came out.
ALTERNATE TUESDAYS STARTING MARCH 5, 12:00-1:00 PM EST on Zoom
RSVP on EventBrite: https://www.eventbrite.ca/e/llm-reading-group-march-5-19-april-2-16-30-may-14-tickets-851921368747?aff=oddtdtcreator
- March 5 The Linear Representation Hypothesis and the Geometry of Large Language Models: https://arxiv.org/abs/2311.03658 with Kiho Park (Google Brain
- March 19 Who Are the Publics Engaging in AI? https://journals.sagepub.com/doi/10.1177/09636625231219853 with Renee Sieber (McGill University)
- April 2 Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models https://arxiv.org/abs/2401.01301 with Matt Dahl (Yale), Verun Magesh; Mirac Suzgun (Stanford)
- April 16 Aya Dataset: An Open-Access Collection for Multilingual Instruction Tuning https://arxiv.org/abs/2402.06619 with Marzieh Fadaee, Ahmet Ustun (Cohere For AI)
- April 30 How Well Can LLMs Negotiate? Negotiation Arena Platform and Analysis https://arxiv.org/abs/2402.05863 with Federico Bianchi (Stanford)
- May 14 Detecting LLM Generated Text in Computing Education: A Comparative Study for ChatGPT Cases: https://arxiv.org/abs/2307.07411 with Michael Liut (University of Toronto)
Join us on Discord for recordings of past sessions, discussions and future session ideas: https://discord.gg/a8auUyKAgB
In partnership with the Human Feedback Foundation: https://humanfeedback.io/
An Anthropologist’s Perspective on Aya
Join Aya French Ambassador, Joseph Wilson, as he explores the cultural dimensions and human implications of the Aya project through his unique perspective as an anthropologist. This talk was given as part of Celebrate Aya, Together – a livestream event to celebrate the release of Aya in February 2024. Learn more about Aya at https://cohere.com/research/aya
Introducing Aya
I am proud to be one of 3,000 humans who built Aya – a new massively multilingual, generative LLM that outperforms existing open-source models and covers 101 different languages. Check out Aya here, including the dataset, model, two new papers on arXiv and a nice documentary on the entire process.
https://cohere.com/research/aya
The paper here details exactly how we put together the dataset and relied on communities of speakers in 101 different languages around the world. Submitted to Association of Computational Linguistics, 2024.
Are Emergent Abilities of Large Language Models a Mirage?
At NeurIPS in December I met Rylan Schaeffer from Stanford, author (with Brando Miranda, and Sanmi Koyejo) of this fascinating paper about the benchmarks used to measure the capabilities of LLMs. He found that many of the most common benchmarks use non-linear or discontinuous metrics to measure capabilities that should really be measured with linear metrics. The non-linear metrics show sudden jumps in ability as models get bigger–so-called emergent abilities. But if you change the metric so it’s linear, as models get bigger they show steady, predictable progress. Nothing magical about it.
Click here for a re-print of an article I wrote for American Scientist, March-April, 2024, Vol. 112.
