6 Astra rewired a small AI internally and raised its math score from 7.9% to 48.1% Researchers with OpenAIs GPT-6 gave Astra the raw numbers inside a small quantum model, its weights, and asked it to improve the math model. No practice run, just edits. Astra found the flaw that caused the small model to get stuck in thought loops, a problem that Quens' own model card warns about. This changed 0.016% of the weights. Even the Israelis of the teaching following the teaching tripled. In other tests Astra tested the raw weights of small game nets and put them to the game each was trained on. He also wrote small game networks from scratch that beat those trained over 100 million steps. The team calls it space-weight programming in a post published today. Wharton Professor Ethan Mullick: "Waiting for replication but it's a pretty big deal. ".

6 Astra rewired a small AI internally and raised its math score from 7.9% to 48.1% Researchers with OpenAIs GPT-6 gave Astra the raw numbers inside a small quantum model, its weights, and asked it to improve the math model. No practice run, just edits. Astra found the flaw that caused the small model to get stuck in thought loops, a problem that Quens' own model card warns about. This changed 0.016% of the weights. Even the Israelis of the teaching following the teaching tripled. In other tests Astra tested the raw weights of small game nets and put them to the game each was trained on. He also wrote small game networks from scratch that beat those trained over 100 million steps. The team calls it space-weight programming in a post published today. Wharton Professor Ethan Mullick: "Waiting for replication but it's a pretty big deal. ".