>>105768174They all suck but Gemini 2.5 Flash isn't terrible as a quick reference at the moment. It was really bad a few months ago. DeepSeek is pretty shit lately, not sure what they did but it just makes up everything. When I tell it how it's wrong it doubles down and assures me it will get it right this time, then it just repeats the wrong thing it said earlier.
Gemini also doubles down when it's wrong but it's wrong less for the questions I ask.
Fundamentally LLMs are a dead end. They are a neat trick and nothing more.