RL-trained Kimi base model designs power transformers, meeting 93% of unseen specs
simonguozirui · x · 2026-09-15
Gentle Trajectory reports using reinforcement learning to teach a Kimi base model to design medium-power transformers (the electrical kind): it met 93% of unseen specs, compressing multi-week engineering efforts into minutes of inference — "we used the transformer to design the transformer."
The thread (1/5) uses this as evidence for a bigger thesis: AI is now improving its own model architectures and chips, so the power infrastructure it runs on deserves equal attention.
Related event: RL-Trained Kimi Designs Power Transformers, Hitting 93% on Unseen Specs(2 posts)→
More from Infra
- Leaker Claims Nvidia RTX Rubin 6090 Launching Next Year — max_paperclips · 2026-09-15
- Pareta routes cheap LLM tasks to small models, 620x cheaper than GPT-5.5 — D33B · 2026-09-15
- Prefill and Decode: why asking an LLM for three takeaways from a long document still takes minutes — dotey · 2026-09-15
- Training from scratch on a single H100 hits 76% on ARC-AGI-1 in ~4 hours — GregKamradt · 2026-09-15
- From Ollama to vLLM: a roadmap for scaling LLM deployment — kalyan_kpl · 2026-09-15
- Kimi K3 is live and free on NVIDIA NIM with OpenAI-compatible API — airesearch12 · 2026-09-15