Artificial Intelligence
Model improvement
Single source · unconfirmed
Prepared by the Prisma AI-based automated newsroom · cross-checked against sources
In early November, researchers from MIT CSAIL conducted an experiment where the GPT-6 Astra model improved the capabilities of the Qwen3.5-0.8B model in passing math tests. This was achieved without traditional retraining, but simply by changing a very small fraction of the model's weights. This approach may open up new possibilities for improving artificial intelligence models.
Sources: N+1
🖼 AI illustration
Sources: N+1
🖼 AI illustration
Prism: who is reporting · 1
Asia & othersN+1
⚑ Spotted an error?
✓ Thank you! The editors will check within 24 hours.