Max Bartolo
@max_nlp
I lead the Command modelling team at @Cohere and co-chair the @DynabenchAI @MLCommons working group. Prev @DeepMind, @MetaAI / FAIR & @BloomsburyAI.
🎤 Meet our expert panelists! Join Albert Gu, Alisa Liu, Kris Cao, Sander Land, and Yuval Pinter as they discuss the Future of Tokenization on July 18 at 3:30 PM at TokShop at #ICML2025.
I’m building a new team at @GoogleDeepMind to work on Open-Ended Discovery! We’re looking for strong Research Scientists and Research Engineers to help us push the frontier of autonomously discovering novel artifacts such as new knowledge, capabilities, or algorithms, in an…
Some of the real-world challenges of building for representation
This is one of my favorite sections in the Aya dataset paper. It is towards the end of the paper, so probably isn't read often. It speaks to how the end breakthrough was completely intertwined with the geo-reality experienced by independent researchers around the world.
NeurIPS is pleased to officially endorse EurIPS, an independently-organized meeting taking place in Copenhagen this year, which will offer researchers an opportunity to additionally present their accepted NeurIPS work in Europe, concurrently with NeurIPS. Read more in our blog…
Really enjoyed discussing the state of AI benchmarking alongside Prof Mark Bishop, @IAmTimNguyen, Enzo Blindow & @ecsquendor at @MLStreetTalk's first in-person event in London yesterday. Looking forward to many more!
LLMs can be programmed by backprop 🔎 In our new preprint, we show they can act as fuzzy program interpreters and databases. After being ‘programmed’ with next-token prediction, they can retrieve, evaluate, and even *compose* programs at test time, without seeing I/O examples.
We’re looking for a Research Engineer / Scientist with a focus on Data Analysis and Evaluation to join the post-training team at Cohere! More details and application here: jobs.ashbyhq.com/cohere/6170371… Feel free to reach out if you'd like to know more!
Looking forward to sharing some of our recent research contributions at @MLStreetTalk's first London AI meetup 🤩
We are running our first physical event in London on 14th July! We have Tim Nguyen @IAmTimNguyen from DeepMind and Max Bartolo @max_nlp from Cohere and Enzo Blindow (VP of Data, Research & Analytics) at @Prolific joining us. Not many seats for the first one.…
Kudos to @cohere for releasing 6 proper research papers in May alone, while publications of other western labs increasingly read like advertisements! I recently read the Command A technical report and it contains much more detail than other model reports. Looking at recent…
the command-a paper is one of my top 5 papers of the year for sure cohere.com/research/paper…
Massive congrats team Afri-Aya, really great work! 🤩
Huge Win Today 🎉🎉 Our team “Afri-Aya” just won this year’s @cohere Aya Expedition. Our work is focusing on 1) curating and evaluating vision dataset then 2) Finetuning the Aya vision model for underrepresented languages in Africa. I represented my beloved Sierra Leone with Krio
Join us to mark the end of Expedition Aya, our six-week global open-build challenge designed to accelerate ML research progress in multilingual, multimodal and efficiency✨ Top teams will present their key findings and innovations and our judges will select 5 winning projects🏆
Congrats to our Cohere colleagues for their paper “Improving Reward Models with Synthetic Critiques” being presented at NAACL this week! 🎉 Read the paper: arxiv.org/pdf/2405.20850 Work led by @Daniella_yz, @FraserGreenlee, @max_nlp, Phil Blunsom, @jaa_campos and @mgalle
Recently overheard @iclr_conf: influence functions for LLMs are useless. Poster #208 disagrees 🤔
At #ICLR2025? Come and see @LauraRuis present these amazing results on how LLMs exploit data in different ways to learn facts vs capabilities. Happening now at poster 208 in Hall 3! 🚀
If you want to learn more about how LLMs pick up reasoning abilities from procedural knowledge in pretraining, visit poster #208 in Hall 3 at 3pm today @iclr_conf #ICLR #ICLR25 #ICLR2025
Presenting this today 3-530 at poster #208, come say hi 🙋♀️
Attending #ICLR2025 and interested in #LLM, #Alignment, or #SelfImprovement? Then come by and check out our work from @cohere: "Self-Improving Robust Preference Optimization" - a new alignment method that unlocks self-refinement in LLMs! 📍 Poster Session 4 — Friday, 3–5:30 PM
🚀 Presenting at #ICLR2025: Self-Improving Robust Preference Optimization (SRPO) — the alignment method powering @cohere flagship models Command A & Command R7B 🔥 SRPO learns to refine itself, outperforming existing DPO, IPO, SLiC & RPO across the board. 📍Poster session 4…
Really enjoyed giving this talk. Thanks for hosting and for the great questions! @tomhosking you might recognise this slide 😅
Another great London Machine Learning Meetup earlier. Many thanks to Max Bartolo (@max_nlp) (researcher at @cohere) for the fascinating talk on 'Building Robust Enterprise-Ready Large Language Models'. And thanks to @ManGroup and @ArcticDB for hosting.