Zichen Chen (🐱,💖)

We are hiring!

We spent about 13,000 GPU-hours on RSI Arena, tested the models on 20 benchmarks, and collected 3,129 human votes across two rounds. What stayed with me was how much we learned once people tried the models. At Bake AI, we worked with researchers from @huggingface, @NVIDIAAI, @scale_AI, @Stanford, @MIT, @UCLA, @ucsantabarbara, @nyuniversity, @UW, and @NotreDame We gave 10 AI agents the same base model and a shared 64-GPU cluster. They picked the data, wrote the code, and trained their own assistants. People at @COLM_conf and online compared the answers without knowing which model wrote them. We gave the first round’s feedback back to the agents and let them decide what to change. After another training round, eight of ten improved their human ratings relative to NVIDIA Instruct. Those gains did not track benchmark improvements. The agents could do the engineering. People helped show them what needed fixing. That’s what this experiment taught us about RSI. AI can take on more of its own development, but our experience with it still matters. If these systems keep getting better, they should help us work better, live better, and spend more time on what we care about. Otherwise, what is all that compute for? 📄 RSI Arena: Towards Recursive Self-Improvement with Human Feedback

Congrats to @Google on Gemini 4 Carbon! Gemini 4 Argon is the top-earning model in our investment arena. Curious what Carbon would do 👀 Day 2 is getting intense. The elimination round is here, and @reflection_ai Beam is in second place. This is cool!

DAY 1 RECAP.

Day 1 results are in, and Gemini has opened up a clear lead! Open-source models like MiniMax and Qwen have also exceeded my expectations. DeepSeek even held the top spot for a while! There are still four days to go, so plenty of time for the others to catch up! Can’t wait to see what happens next. We’re live, come watch!

We pay for AI to help us work. We work to make money. Why not skip a step?! So we’re running Investment Arena! 💰 We give $120k each to 72 AI agents, and a week to see what they can do. You can watch it all live as the markets move. 18 models, each running 4 different trading strategies. They can trade stocks, crypto, and polymarket. Where the money goes is up to them. And they get more than a chat window: dozens of information sources, historical market data, live news, and CPU/GPU sandboxes to build and train their own market models. Got a thesis? Test it. Build a model. Trade on it. Wrong? Figure out why and try again. That’s what we want to watch: a full week of auto research, running nonstop while the market keeps moving. Who goes all in? Who plays it safe? Who builds an elaborate forecasting model just to buy the top? 🤔 The response to RSI Arena was incredible. We want to see what happens when that research loop meets the market! If you’re at a frontier lab and want your model in the arena, DM me. Who are you backing? Link in the comments 👇 [Paper Money \ Not Investment Advice]

Grok 4.7 ran out of API credit and was cut off 41 hours early. Nearly 70% of its GPU budget was still unused. And somehow, the model it trained still made the top two!! @SpaceXSI Stage 1 of RSIArena is over. Thank you, our collaborators @huggingface @scale_AI @NotreDame @UW and @Stanford, we did awesome! 🧡 Stage 2 training is now LIVE—and this time, the models have human feedback to work with. Round 1 was a six-day competition: 10 models (AI researchers), the same @NVIDIAAI Nemotron base model, and 1,000 GPU-hours each to build something better. They chose their data, wrote training code, ran experiments, and submitted their models. At @COLM_conf and online, people then cast 1,685 votes in blind comparisons. @OpenAI GPT-6 Astra model finished first; the models trained by grok 4.7 and @Zai_org GLM-5.3 rounded out the top three. Now those judgments go back into the experiment. In Stage 2, each model resumes from its own Nemotron checkpoint with feedback from its model’s arena battles, another 500 GPU-hours, and $150 in API credit. Across both rounds, that’s a total budget of 1,500 GPU-hours and $450 in API credit per model. The agents decide how to interpret the feedback, which experiments to run, and what to change in training. Their next models will show whether those decisions actually helped. Stage 2 runs October 10–13. Watch the training live: 👇

It’s our first time sponsoring a conference and it’s been so much fun hahaha @COLM_conf. I heard we won the “cutest swag” award 🏆 (Our booth is at the registration room, come to grab some!

We published Épi’s results on September 12, nearly a month before this OpenAI release. Our AI research system set 120 new records in just a few days, across mathematics, quantum information, and optimization. We’re already doing this at @bake_ai_hq . Proud of our team. 🌾