Comments by "" (@retagainez) on "Why I Quit Copilot | Prime Reacts" video.

  1. Also unfortunately, if you use it any specific use-case (especially one where a common design pattern in a common language isn't present) it hallucinates on every, single, prompt.
    3
  2. @pnk4996 I'm confused on how you're disagreeing with OP. The LLM might have a different response based on the most common test case for your scenario, especially when the context of your prompt is niche and not in its training data. Test cases aren't for mimicking other people's tests. They're for writing design that you would use as if you were the end user, with an added side effect of having a way to track progress. It's text generation that's translated from a plain text prompt.
    2
  3.  @-Jason-L  And LLMs are terrible at TDD.
    2
  4. I think creative writers are getting the largest bang for their buck with all this LLM business. If you write code as if you were a writer it could really create some creative ideas (even if they're completely wrong in the context of your codebase) and help you brainstorm as a result. It's a lot harder to mess up creative writing than code since it can be more subjective. That's not to say that LLMs don't regularly mess up writing stories by overusing cliches, having wonky plots, or nonsense/lifeless characters.
    1
  5. I'm honestly unsurprised Prime has quit Copilot. He is all about curiosity, but also staying on trendy trends. It's his job after all, especially so with him leaving Netflix. There are suggestions that LLM generated code is worse than junior level code. For the price, it's not worth it. Particularly when they start reducing/rate limiting responses and the quality of those responses.
    1
  6. Most definitely. you need to be able to speak in its language. I like to use it for exploring potential design patterns and interactions between design patterns.
    1
  7. Depends on the physics videos. If it's casual and in layman's terms, probably. But if you're watching lectures and exercising your mind along with the lectures that's irreplaceable and isn't easily done by LLMs. I'd argue doing fervent research is a lot different from fervently prompting an LLM for information/code which may or may not be hallucinated. I'd also argue researching with physics videos would supplement your ability to prompt AIs about physics questions in a way that's easily translatable to other physicists. Many things in moderation perhaps.
    1
  8. People depend on it to write working code, but somehow manage to avoid the in-depth learning they could achieve from doing cursory searches on the topic they're working in. Great research tool to an extent if you can express yourself sufficiently in a single prompt.
    1
  9. If it's a niche bug that requires context, it will be terrible for that. But if it's a bug based on a flawed design pattern, it might do well in pointing that kind of bug out.
    1