Comment on If AI was all it was cracked up to be, it wouldn't be shoved in your face 24/7

<- View Parent
SaraTonin@lemmy.world ⁨12⁩ ⁨hours⁩ ago

They’re useful, but not anything like to the degree that’s being claimed. It’s being pushed as if it’s going to be able to do everything.

For one example, does generative AI have a use in coding? Absolutely. If you’re a coder who knows what they are doing, it can help you have ideas and it can do some of the tedious stuff. But you still need to know what you’re doing, use it as a tool, and absolutely do not use its code without checking it all. And that’s not how it’s being pushed. It’s being pushed as if you can just type in “pretend it’s 1999 and code me a Doom sequel” and you’ll get a full, working programme out.

Even getting it to do things in chunks seems to be a trial. There’s a video I watched a day or two ago (if you’re interested, I’ll look for a link when I’ve got more time) where a guy tried to get ChatGPT to code for him. He had a specific end goal in mind and asked it to do things step by step. While there were several occasions that he was impressed by what it produced, he kept getting stuck because it would seemingly only fix problems in one area of the code by eliminating another area entirely. He found that most of his time was spent going round in circles trying to ensure that it actually did what he was asking of it. It hallucinated, too, and at one point he had to solve the problem for it by referring to a specific repository, which he said he himself only knew about because he’s got more than a decade’s experience.

That’s the biggest issue - what LLMs are capable of is far, far below what they’re sold as.

source
Sort:hotnewtop