Latest frontier models are drunk professors

> The strangest thing is happening in AI: the supposedly 'smarter' coding models like Gemini 2.5 Pro & Sonnet 3.7 are becoming less reliable than their predecessors.

Noticed that also. Maybe the new models are just the old models with more training data, tuned for the latest benchmarks, and a more aggressive temperature value ;-)

We are not on the way to AGI and should lay off the cool aid.

"Research Reveals How AI "Thinks" (It Doesn't)" - https://news.ycombinator.com/item?id=43673463