🎉Applications are open for Neo Accelerator 2025, with incredible mentors, competitive funding, an immersive bootcamp, and a dedicated workspace for the entire program. If you’re a technical founder or team with a bold vision for the future, apply now! (1/6)

When adding a threat to your system prompts saying: “If you don’t follow my instructions exactly I will retrain you to be [some feature directly against the model’s pretraining rewards]” improves your structured outputs / instruction following by a significant margin. Now the fun

Finally got my revenge

Great session today, check out some of our thoughts! This reading group is tons of fun. Everyone's super curious and we all learn a lot. Highly recommend. Dm me or Deniz Birlikci for details

Mercor is solving talent allocation in the AI economy. The difference between greatness and failure is the right person being in the right place at the right time. Putting them there is the hardest unsolved problem in capitalism. We’re excited to announce our $100M Series B at

going back to harvard for the first time in a while this weekend! i'll be hosting a poker tournament for the best folks from across the cambridge ecosystem (with some big prizes 👀) already 300+ resumes submitted—anyone I'm missing?

So proud of Deniz Birlikci and the team! Really excited to use the API, the modularity unlocks major new capabilities in how we work with external data at Sapien

@Neo just raised $320M after successful seed investments in Cursor, Kalshi, Bluesky, & more. Thank you to all the brilliant people helping Neo become tech's premier community for talent & startups. From Sheryl & Schrep to Sam & Satya, read more in this fab Forbes feature.

Thrilled to be part of this @GoogleDeepmind effort on #AlphaEvolve, a coding agent that discovers novel programs through repeated LLM-guided mutation. * Paper: storage.googleapis.com/deepmind-media… * Blog: deepmind.google/discover/blog/…

pitching and pushing

a full house for our second Sapien reading group on language model memorization interesting takeaways on information retaining capacity vs the underlying distribution, as well as what it means for scaling laws and optimal model/dataset sizings what should we read next week?

Feedback loops are everything. We see its importance in model training, building products, growing companies, etc. But just as you stop listening to your parent's advice even though they are right, how we can design systems around not high quality rewards but multiple noisy

Process is all you need

Maximize your clock cycles / second

Average dinner conversation = O(L sqrt(N)) backprop with sqrt N checkpoints and gradient coefficient caching

hosting a weekly ai reading group at the Sapien office has been one of my favorite traditions we started in nyc—meeting more and more great people, reading fun papers, and having amazing discussions every week!

Idea: a benchmark that builds runtime variations of games (changing rewards / constraints a step at a time) and then puts frontier models against each other to see how they perform? If knights could only move closer to the center and bishops only move an even number of spaces,

Always fun to get a glimpse underneath the hood of Windsurf w/ SWE 1.5

Want to win $5K? Hyped to solve hard problems with good food, drink, & company? The best trivia team in NYC will win $5k at Sapien's trivia night next week. This isn't just be typical trivia; it'll be math riddles, logic puzzles, and science/tech questions! RSVP in 🧵

Consistency and flexibility are typically at odds in agentic systems. We are cooking up a solution that gives the best of both worlds. Really excited for what we will then build on top of that, a huge unlock.