There's a buzzy new paper about GPT-6 Astra being able to directly interpret model weights (prompts full of floating point numbers) - I tried it with Qwen 3.8 27B on my laptop and found that, after 45+ minutes, that model could achieve the same thing https://simonwillison.net/2026/Nov/3/qwen-can-interpret-model-weights/
Simon Willison replicates GPT-6 Astra weight interpretation with Qwen 3.8 27B
AISimon Willison says a new paper claims GPT-6 Astra can interpret model weights directly from prompts of floating point numbers. He ran the same test on Qwen 3.8 27B on his laptop and says the model achieved the same result after more than 45 minutes.
Simon WillisonVerified on X
@simonw
Source details
Simon Willison · x.comPublished · added here
Original title: There's a buzzy new paper about GPT-6 Astra being able to directly interpret model weights (prompts full of floating point numbers) - I t...