An MIT-developed frontier LLM agent, GPT-6 Astra, repaired reasoning loops in Qwen3.5-0.8B without gradients using weight-space programming, according to work by Takehi et al., consuming 158 agent turns, 89.7 million tokens and 50 forward passes. The technique, described as Tree-Ising bit-level editing, lets an agent read a network's weights and write a repair directly.
Weight-space programming, introduced in recent MIT work by Takehi et al., enables a frontier LLM agent to read a network's weights and write a repair. Their GPT-6 Astra agent corrected the reasoning loops of Qwen3.5-0.8B without gradients, using 158 agent turns, 89.7 million tokens, and 50 forward-p
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.