4 dispatches tagged autoresearch
Yes! An autoresearch loop of 81 experiments ($364) found a special-purpose neural memory architecture that ‘remembers’ an exact geo-location from an in-flight daytime picture of Berlin — typical miss 27 m, no GPS, no internet, no map on board, just a 3.1 MB file of weights that works on 96.5% of camera frames.
An autoresearch loop — a genetic algorithm over fourteen genes of the real, open-source Source One V6 plate drawings, flown through six simulated weather scenarios, with Claude sitting in every few generations to propose designs from the run’s own telemetry — evolved a 7-inch frame that flies 16% more efficiently (Wh/km) than the stock champion.
Autoresearch on a job-shop scheduling problem dressed up as a lunch rush: one grill, one shared range, sequencing as the only lever, and an agent that proposes candidate rules, runs them across simulated services, and keeps whatever beats the champion — trained on six scenarios, then held out against three more to check it found a principle, not a memory. Built on Nima H. Siboni’s “The Heuristic Scientist: Open-Ended Algorithm Discovery with LLMs” workshop.
Autoresearch — 96 experiments across every lever we could find: prompting, retrieval, decoding, pipeline, fine-tuning, preference-learning, reinforcement-learning, adversarial-distillation — found a technique that brings Apple’s native 2025 3B Foundation Model within about one rubric point of a frontier cloud model (Claude Sonnet 4.6) on every quality dimension.