Flint 40B self-improves into Flint-Law, gaining 15% on legal tasks over 11 steps
RichardSocher · x · 2026-11-15
A user experiment retweeted by Richard Socher prompted You.com's Flint to "make yourself better at Law," evolving Flint 40B into Flint-Law 42B-a3.4b via 11 self-improvement steps.
Results: +15% average score vs stock (0.473 → 0.544), improvements in 7 of 8 task categories, with the biggest gain in compliance certificates (+30%, +0.132). Deficiencies detection was the only decline (-4%). The author plans to run another day and post the checkpoint.
More from Research
- Sebastian Raschka publishes from-scratch RoPE tutorial and survey of positional embeddings in modern LLMs — rasbt · 2026-11-15
- Training your own LLM judge for ~$5 in 1h40 on one GPU, 89.5% accuracy — MaziyarPanahi · 2026-11-15
- Ivy Grad Trading Analogy: Reading More Doesn't Help Without Deliberate Practice — minhnguyen · 2026-11-15
- LeCun's CoRL slide: 'If you are interested in human-level AI, don't work on LLMs' — Heavy-Ad9232 · 2026-11-15
- Dataset Distillation via Drifting compresses 1,300 ImageNet images per class into just 10 — kastnerkyle · 2026-11-15
- "Double-blind review doesn't work": reviewers see author names, and rival students reject competitors — Temporary_Switch_339 · 2026-11-15