Apple showed models can self-improve at coding through simple self-distillation — sampling their own outputs and fine-tuning on them with no verifier, no teacher and no reinforcement learning — lifting one model from 42.4% to 55.3% on LiveCodeBench, with gains concentrated on the hardest problems.
Meta translated an entire graduate mathematics textbook into Lean using 30,000 agents, turning proof into a parallelizable compute job.