🇨🇦🥲

DatologyAI and I are at #ICML2025, and we have two best-in-class things at the conference: 1. Swag: fidget spinners 2. Jobs: We're looking for talented researchers and engineers (full-time & interns). We're especially interested in someone to lead research on data curation for

We are looking for a post-training lead at DatologyAI we have gpus, you can make them go brrrr

I've learned from many writer heroes of mine on Dialectic with Jackson Dahl. Humbled and excited to be able to share this wide ranging conversation about agency, technology, interfaces, and what I've been working on lately with the inimitable Jackson Dahl 💙

I just wrote my first blog post, “Everything is Training Data Now.” I noticed myself starting to doubt LLM performance on certain benchmarks due to data contamination / gaming, and wanted to see how real this is. Spoiler alert: there’s a lot of variance across models and labs.

👁️👌

come hang out on Oct 9! pizza & drinks brought to you by the incredible Plymouth team; talks from Ahmed Ahres, Niccolo Fusai, and myself

meticulous anti-consensus take

Trying to raise one last batch of funds for season 1 of Toronto School of Foundation Modelling! We are nearing the end of the first season with many in the cohort already training their own models, discovering scaling laws, building tokenizers etc. I’m planning a few surprises

🙌Biomni-R0-32B, the first RL agent model trained end-to-end for biology research, is now open-weight on Hugging Face! To benchmark progress, we’re also releasing Biomni-Eval1 — 443 data points across 10 biomedical research tasks! 🔗 Data: huggingface.co/datasets/biomn… 🔗 Model:

certain words are so nuanced that their translation is an impossible task lexiconic is a digital exploration of those words basecase.vc/lexiconic

We are hiring a data adaptation research engineer. 🔥 This is a role I am really excited about. Our hiring says a lot about what we stand for, and how we want to build technology. So, here we go...

learning all about AI for science, the future of robotics, and localization from the brilliant Ahmed Ahres, Nicole Fitzgerald, and Niccolo Fusai at Plymouth HQ. Lessons learned from building Physical Intelligence Revolut and Mila - Institut québécois d'IA 💙

have been meaning to write an overview on this exact topic but Corin's already done it beautifully

I am specifically looking for a talented early career researcher (or engineer looking to break into research)! - Fast ramp-up to work on frontier capability evaluation at pre-training & post-training stages - Abundant technical mentorship and exposure to major model labs -

Happily overwhelmed by the response to this! Addendum: am also looking for a very special type of mid-career researcher, more below: - Opportunity to stretch and grow as a leader, a team-builder, and a setter of research direction - Again, ample technical and personal

am working through responses to this! will get back to everyone within the next 48 hrs

I've got something new for everyone. My first substack article! Not the one I planned to do first, but a fun one! I have made a handy calculator base on the DeepSeek v1 coefficients for finding optimal LR and batch sizes for dense LLMs.