| I'm more bemused than burnt out... kinda surprised the hype around it doesn't get called out more often by developers. LLMs as coding assistants are undeniably time saving devices, especially when working in languages/libraries/platforms/frameworks you aren't already very familiar with, or when needing to generate something very boilerplatey as a one-off. I am not calling the technology useless by any stretch of the imagination, but its still just so wildly overhyped right now. It is a pretty common occurrence these days for me to have a blog post open from some "AI industry thought leader" talking about how all developers will be out of work in a year while at the same time I have a Gemini window open and I'm just watching it absolutely flail on relatively simple things like generating a database query or a regex (that is novel and not something that's scattered all over its training set like a simple email validator). And Gemini 2.5 is, IMO, the best of the models when it comes to programming assistance (having replaced Claude 3.5 which was IMO previously the best), at least for the areas I touch (lots of kotlin/KMP/Android/etc). As goofy as Gemini sometimes gets it is far less frustrating than asking Claude 4 a question and watching it write out a whole ass answer but then correct itself like 7 times before finally coming to a shitty answer that is worse than 2 of the ones it wiped out while blowing through most of its context window on its loop of indecisiveness. And relatedly... color me completely unsurprised that this thread got dumpstered off the front page so quickly. Gotta keep pretending like the singularity is going to happen next week. :D |