tl;dr: AI researcher (EBM), $225k–$350k+ plus equity, San Francisco

So, we’re hiring researchers for a startup building reasoning AI. Which is not uncommon per se, but here’s the situation:

• Founding Chair — Yann LeCun, Turing Award laureate and godfather of AI
• Chief Mathematician — Michael Freedman, Fields Medalist
• Team — 10 PhDs and six ICPC medalists on the team
• Founder — A quantum physicist from Berkeley who wrote her PhD under the 2025 Nobel laureate.

The reasoning AI in question is not LLM-based, not even language-based at all. It's EBMs — energy-based models. Instead of predicting the next token, they minimize an energy function in latent space. High energy means something's wrong; low energy means you're close to the truth. Lots of people once believed reasoning models would work exactly this way — including LeCun himself, who's been pushing for EBM since the 1980s.

What we got instead was simply RL on long chains of thought. Not a bad outcome — but a compute-costly one for sure. EBMs optimize not for likelihood the way LLMs do, but for correctness — "what minimally violates the constraints." And rather than generating tokens one by one, they optimize the entire trace at once — with the ability to refine it iteratively 🔧

One of many applications: writing formally verifiable code far more effectively than LLMs can. Which means reliable systems for pacemakers, financial markets, nuclear reactors — you name it.

Logical Intelligence works on both EBM development and formal verification. In one benchmark, their model solves 96% of hard Sudoku puzzles — while frontier LLMs score around 2%. Their formal verification agent scored an insane 99.4% on PutnamBench — and corrected 15 errors in the problem set itself along the way 🔍

We're looking for an AI Researcher with a fairly specific profile:

• MSc / PhD;
• publications at ICLR, ICML, NeurIPS, or CVPR;
• ideally — experience and publications in EBM;
• but strong backgrounds in diffusion models, LLM fine-tuning for reasoning, non-autoregressive reasoning, or MCMC in latent space may also be a fit.

Details:

• $225k–$350k+ plus equity
• On-site in San Francisco
• O-1 visa support available
• Reach out to Fedya 💜