The Decoder· Maximilian Schreiner·· 3 h ago
GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights
GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights
AI summary
MIT researchers show that OpenAI's GPT-6 Astra can read and deliberately rewrite the weights of other AI models without traditional training. In one experiment, Astra boosted a Qwen model's math accuracy from 7.9 to 48.1 percent by changing just 0.016 percent of its weights. The approach could fundamentally simplify AI optimization but still hits clear limits.
Selection record
Threshold 76Media and individualsFirst 82Second 88
AdmittedSum of both 170 ≥ twice the threshold 152
- Source tier
- Media and individuals; this tier's threshold is 76
- Pre-filter
- passed：GPT-6直接改写模型权重，实质AI技术
- Why it was chosen
- Shows a language model reading and writing network weights directly to diagnose and modify model behaviour, with concrete gains on MATH-500.
- Same event
A model scores each item twice, independently, against one written standard, out of 100. An item is admitted only when the two scores add up to twice the threshold. The threshold is set per source tier.
Source: The Decoder · the-decoder.com