Flint cuts bike-demand validation RMSLE 30.8% in 3 hours of autonomous model iteration
cong_ml · x · 2026-10-22
- Task: predict bike rentals from weather and calendar using the public UCI Bike Sharing dataset.
- Flint autonomously combined log + quadratic time trends with residual trees; the chart shows best validation RMSLE dropping from 0.537 to 0.3708 within 3 model hours, via steps like separating trend from trees, switching the trend model from Poisson to Tweedie, and combining time scales.
- Final results: 30.8% lower validation RMSLE and 9.1% lower held-out RMSLE, demonstrating practical autonomous feature engineering and model selection on a local model.
Related event: Recursive Flint runs autonomous research on a laptop, open-sources 40B-A2B(7 posts)→
More from Models
- YC F26 startup Latent reads model internals to catch hallucinations at 530x lower cost — ycombinator · 2026-10-22
- Perplexity's JOLT matches GRPO with 6.5x fewer training tokens, JOLT+ wins on all five benchmarks — perplexity_ai · 2026-10-22
- Recursive AI ships Flint, a 40B-A2.5B MoE claimed to match Qwen3.6-27B, weights on Hugging Face — ChengleiSi · 2026-10-22
- DeepSeek ships V4.1-Flash-1017, now the #1 open model on RSI-Inference Bench — scaling01 · 2026-10-22
- OpenAI's Cybersecurity Guardrails Block a Dev From Auditing His Own Patched Server — Next_Technology6361 · 2026-10-22
- TPN-006 hits Hugging Face: 7.7GB compressed model beats its 17GB Mistral teacher — markjeffrey · 2026-10-22