News for hackers. Stuff that matters.
Re:Effective AI code suggestions: less is more by GavCo on Wednesday January 29, 2025 @03:11PM (#42870483)
Interesting. I wonder if this is related to the model architecture and attention mechanism.

The author seems to be implying it could be: "Even a single mention of ‘code enhancement suggestions’ in our instructions seemed to hijack the model’s attention"

Re:Effective AI code suggestions: less is more by jimminyx on Wednesday January 29, 2025 @03:45PM (#42870930)
The attention is probably just latching on to strong statistical patterns. Obvious errors create sharp spikes in attention weights, and drown out more subtle signals that can actually matter more