I've had the most current publicly available models fail to give me even simple correct boilerplate code, but the guy is like: ...we have to be nuanced but, "Co

I've had the most current publicly available models fail to give me even simple correct boilerplate code, but the guy is like: ...we have to be nuanced but, "Converting dozens of programs to C or Rust to improve performance 10-100x."? Seriously? I also recently asked openai's gpt 4 which number is bigger, 3.8 oder 3.11, and it was pretty damn sure that it's 3.11, because 11 bigger 8, obviously. Another time I asked Meta Llama 3.1 70B and gpt 4 multiple times using a variation of prompts to suggest a simple code example for a feature in a project I was working on. They confidently provided code that was total nonsense, configuration that did nothing and even offered a dependency that didn't exist but somewhat sounded like the prompt itself. I cannot predict the future. Maybe all of this will lead to something useful. Currently though? Massively overhyped. I've talked to CS colleagues and friends who also work as software developers that are all way more competent than me about their experiences, and some were exited about the prospects, none could provide a current use case for their work. The only instances I know in which people talk positively about these models are in online articles like this or in comment sections adjacent to them. Never in real live among actual developers in person.