AI Research Foundry London: Data Is All You Need
AI Research Foundry London is a paper club for AI researchers. Free forever. It's a community not a sales pitch.
Session 01: Data Is All You Need
One theme: data. Creation, curation and scaling.
In September, Dwarkesh Patel and Jerry Han published "Pretraining progress is mostly coming from data". It started a conversation about how much data drives AI progress. We are starting the community right there.
Speakers
Joshua Placidi - Imperial College London (joshua-placidi)
Highlights
PhD student in Imperial College London's Brain and Behaviour Lab, supervised by Aldo Faisal and Marek Rei.
Researches scaling self-supervised learning across medical data with generative foundation models.
Previously a Senior MLE developing AI for scientific applications. Also building Nightingale AI.
Talk: Sequence Composition in EHR Foundation Model Pretraining
Yash Bhalgat - Google DeepMind (yashbhalgat)
Highlights
Senior Research Scientist at Google DeepMind, working on Gemini Omni.
Lead captain for Omni 1.5 pre-training.
Previously at Qualcomm AI Research. DPhil in Computer Vision and ML from Oxford.
Talk: [TALK TOPIC TO BE CONFIRMED]
Maximilian Croci - Microsoft AI (maximilian-croci-6040b485)
Highlights
Works on pretraining data for MAI Thinking models.
Co-lead on the web data effort for MAI-Thinking-1.
Focus on math, science and code from the web.
Talk: MAI-Thinking-1: Building a Hill-Climbing Machine
Pierre-Carl Langlais - Pleias (pierre-carl-langlais-b0105b10)
Highlights
Co-founder and CTO of Pleias, a French-German AI lab for data research.
Released Common Corpus, an open multilingual pretraining dataset of around 2 trillion tokens, and SYNTH.
Work reused by Anthropic, Nvidia and StepFun.
Talk: It's All Training: A Fully Synthetic Single-Stage Recipe for LLMs
Who it's for
AI researchers in academia, big tech and startups. People who want to dig into how AI systems work, how to build them better and how to understand them.
How it works
We start with a paper discussion, then open it up. Read the paper if you can. We share links to summaries before the session.
Why
London has communities for builders. It has nothing for researchers to go deep on one shared idea. So we are making it.
Seminar room energy. Lots of curiosity. A hunger to push the frontier.