Local auto-research with Flint: 128 experiments push circle packing to 2.6294
cong_ml · x · 2026-10-22
congml introduces Recursive Flint, an automated research system that runs entirely on your own laptop, demonstrated on the classic circle packing problem.
- The task: fit circles in a square and maximize total radius; Flint iterates by writing code to move/resize circles, polish layouts, and try new starts
- Per the chart, 128 experiments over 7 model-hours: staggered hexagonal layout, multi-start SLSQP with feasibility checks, 15° start + linear program — lifting the mean radius sum from 2.17 to 2.6294
- The author notes it's surprisingly good for a model working purely from code and scores, with no vision
Context: the wave from OpenAI's Navier–Stokes work to AlphaEvolve, now running locally.
More from Research
- Python array API standard unifies NumPy-like operations across libraries — GaelVaroquaux · 2026-10-22
- scikit-learn's CPU-bound dependencies have become a decade-old bottleneck — GaelVaroquaux · 2026-10-22
- JHU Paper Shows Aggregate Benchmark Scores Can Overstate Model Competence — DanielKhashabi · 2026-10-22
- Megakernel framework claims trillion-parameter pretraining in a single kernel — basedjensen · 2026-10-22
- Pinal on Nature: Designing Functional Proteins Directly from Natural Language — bravo_abad · 2026-10-22
- 136-Country, 118M-Article Dataset Audits State Media Share in LLM Training Corpora — deliprao · 2026-10-22