А вот я думаю что анализ кода для ии не такая уж сложная задача, особенно если ии будет иметь возможность исполнять отдельные куски кода, манипулировать входящими и отслеживать исходящее. Анализ кода даже проще должен быть чем анализ художественного текста.
Disagree. Try the Gemini line of models from Google, they are incredibly good at writing literary lines in English. It's not even close to OpenAI's GPT or whatever. I read a good amount of literature every now and then and I've had jaw dropping moments with Gemini in terms of it's ability to magically form sentences. It sucks at coding though.
The only part I agree with is the execution capability, it's a major superpower to these godless LLMs but literature analysis is much more close to reality than code. Ask Gemini (not GPT) about various interpretations of Cain and Abel and to play the devil's advocate against itself and all sorts of weird things, take it to the extremes, I bet you'd be surprised the more you play with it. Code, not so much. But yeah, their coding capabilities are definitely impressive, I'm not being dismissive of them.
Gemini, the paid ones.