#burnt out on blue fox
1 messages · Page 1 of 1 (latest)
I'm leaning towards ludicrous
Or rather just very, very expensive
This seems like something that will be more or less made available in the short-term future. Yes, if you want to train and host your own model you can, the best OS models are "okay". There are people around that mess with them. #💬general doesn't get a ton of traffic but you can scan through to find some of the people talking about this @dreamy kernel
You're going to run into serious, high risk problems dealing with hallucinations. First question you're already fucked on, web search is NOT robust and won't be for a while
That said, people are using LLMs for this purpose now, just double and triple check the results by hand and assume it's not exhaustive
I have gotten it to hallucinate basic physical properties of not-uncommon substances like methylene blue
I mean, research is a task where humans already fuck this up constantly through their lack of diligence, and LLMs can do the same but express even more unwarranted confidence. and they don't feel bad about lying
Um, I mean I forgot the name, someone dropped a recent launch of something trying to help with research discovery. it's in one of my hundreds of open tabs. Not looking but it also shouldn't be that hard to find. Don't reinvent the wheel, especially when people with a much bigger budget than you are polishing the wheel as we speak
lmfao what
not the name, that sounds stupid expensive, emphasis on the stupid
I don't know anything about it but it's very difficult for me to imagine anything worth that
oh, fucking proprietary data then
there we go
oh, weak. But I mean, they charge what they can get away with 🤷
I think China just open sourced a protein folding-related AI tool to compete with some proprietary shit. lol
Which good, I hope they snap the spines of the people hoarding these models
you are very optimistic. maybe next after next
Admittedly, I don't know the state of the art. not my field anymore
but I do think there's a little irrational exuberance just because these models seem really smart at first glance
They're really smart until they're not lol