David HodgeWhat's #rstats thoughts on LLM's and neural networks (i.e. "AI")? Underhyped or overhyped?
ahistorical immaterialist@davidhodge931 overhyped, environmentally destructive, very limited use compared to the money and promotion being wasted on them
@jimbob Yeah, incredible amounts of cash and energy used in these.. But flexible useful tools?
@davidhodge931 honestly I feel the supervision and mental effort you have to put in, as a user, to make sure they haven't made a mistake, negates much of the advantage. In terms of using them for generating code, you can't just ask them to write code, you also have to carefully check, understand and test the code, and in that time I could have written it myself.
I *have* made them do useful things, but about once a month - I just can't get my head around people using them constantly, using them enough to make a paid subscription worth the cost.
Tobias Zimmermann@davidhodge931 Overhyped, in my opinion. Are there settings in which the tech can be useful? Certainly. However, since many things AI are still being advertised and sold as silver bullets, they are set up for disappointment. I am currently working on a scoping review about AI research in the medical field and it's quite harrowing what 'science' has been produced by throwing AI at everything.
tuban_muzuruAn LLM is useful for a few things, but if it's backstopped with a neural net, trained in an area of specialty - overhyped. It's as good as the training data.
Anyone still using the acronym AI is trying to sell you something.
Kathe Todd-Brown@davidhodge931 LLM, CNN, and other AI are use case dependent. If you need them, they can do things no other method can. Their use case is narrower than most people are trying to apply them. Then they suck.
vim exiter@davidhodge931 I don't speak for the whole community, of course. I was a skeptic, then I started using copilot at work because
1. my co. paid for it and accepted the risk and 2. it worked in vim. I wasn't sure what to think, until I had to reimplement some tidyverse code in pandas (🤮). It ... just did it, saving me so much web searching. I'm not sure I'm faster now because I have to check everything, but it saves some typing, which isn't nothing.
1. my co. paid for it and accepted the risk and 2. it worked in vim. I wasn't sure what to think, until I had to reimplement some tidyverse code in pandas (🤮). It ... just did it, saving me so much web searching. I'm not sure I'm faster now because I have to check everything, but it saves some typing, which isn't nothing.
Alexis Weinreb@davidhodge931 I think it's part of the future and is a great tool for some problems, but definitely overhyped.
Cameron Mulder@davidhodge931 skeptical and very concerned about what use does to critical thinking and de-skilling ( see https://advait.org/files/lee_2025_ai_critical_thinking_survey.pdf) but I have found LLMs useful for helping me write email and put together initial drafts of documents.
Lorenzo Isella@davidhodge931 I think this technology is here to stay for better or worse. I like LLM for polishing text which I wrote and they save me a lot of time with coding tasks of average difficulty,but I have been disappointed every time I asked an advanced coding question. However, if I were now to learn coding or maths and I used extensively LLM s, I think they would do more harm than good to my understanding of any language and task in general.