Simon Willison Says Free Qwen Opens AI Claims To More Checks
Simon Willison said Qwen 3.8 27B, a free AI model he ran on his laptop, interpreted another model's raw weights after about 45 minutes of setup, lowering the barrier to outside scrutiny. Key Takeaways Simon Willison said Qwen 3.8 27B running on his laptop could interpret raw model weights Willison reproduced one of the paper’s ideas locally after about 45 minutes of setup The paper said GPT-6…
Simon Willison, an independent developer and researcher, demonstrated that his Qwen 3.8 27B model running on his laptop could interpret raw weights from another model after about 45 minutes of setup. This development lowers the barrier for outside scrutiny and replication of AI model claims. The paper behind this experiment suggested that GPT-6 Astra could edit 0.016% of Qwen3.5-0.8B's weights, improving MATH-500 from 7.9% to 48.1%.
Willison's laptop test broadens the field by making weight space programming more accessible, allowing more independent researchers to test the technique on various architectures and real-world tasks. However, the strongest parts of the paper still rely on controlled setups, and safety risks are apparent. If more models can read and edit weights, it could change debugging, auditing, and patching processes, moving away from traditional retraining and fine-tuning methods.
The next step involves replication tests to determine if weight space programming leads to reliable improvements or remains limited to specific demonstrations.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
