I tried to make it compare features from different products and they were making lots of mistakes.<p>Anyone is getting the same impression that AI models are worse now than before?
Gemini for sure, in my experience. I asked 2.5 (presumably the most advanced in the known Universe) to write some code. First it came back with suggestions on HOW to write that code. Asked again. He wrote some, but from 4 functions implemented, 2 were just comments explaining again how to implement that function. Notable mention, one of them was supposed to call the Gemini API. Everything Python, nothing fancy. Took me like 6 rounds of back and forth until I got some very basic structure... So, no vibe coding with Gemini for me.