GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights

By Maximilian SchreinerSource:

[The Decoder](https://the-decoder.com)
        MIT researchers show that [OpenAI](https://www.machinebrief.com/glossary/openai)'s [GPT](https://www.machinebrief.com/glossary/gpt)-6 Astra can read and deliberately rewrite the weights of other AI models without traditional [training](https://www.machinebrief.com/glossary/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](https://www.machinebrief.com/glossary/optimization) but still hits clear limits.

The article GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights appeared first on The Decoder.

Get AI news in your inbox

Daily digest of what matters in AI.