SETA got accepted to NeurIPS 2026! 🎉 We released the full training scripts for DeepSeek-V4-Flash, GLM-5.2 (LoRA), Inkling-Small, and Qwen3.8-27B, with both GRPO and PPO, alongside 4,500+ open-source, verifiable terminal RL environments Huge thanks to the @radixark Miles team for helping us scale the experiments!

why can the china labs build glm-5.2, kimi k3, and many more to come? it is because of the openness. not just the open weights but the whole ecosystem. most of the work done in the china labs is carried by interns. i met brilliant undergrad and graduate interns who deeply understand the model training details, and they are 100x more open to share. that means the talent that knows how to train llms in china is 100x greater in number than the talent in the us, and it is growing in contrast, the us ai ecosystem is too closed. frontier labs do not hire interns. i know brilliant phd students at stanford, berkeley, and so on. they struggle to get an internship and the compute to train a properly sized model. most of the secret recipes are locked away by a very small group of privileged researchers it is not about china or the us. it is about open and closed science. the fact is that every average cs student can learn how to train an llm. they just need the opportunity. labs should be more open and hire more interns, like how deepmind and fair did in the pre-llm era

"behavioral state decay": the failure mode that long-horizon agents forget what matters. meta ai's fix: a memory agent that updates the memory bank and then decides whether to emit a proactive intervention. they train qwen3.5-27b on seta using sft and grpo and show gains on terminal-bench 2.0. great to see seta terminal agent rl envs used this way - remember when it matters by meta ai @yifannnwu @zhuokaiz: - our seta project: