A reasonable summary of the field (a technologist's eye view).
-
RE: https://mastodon.nl/@bert_hubert/116000388750823868
A reasonable summary of the field (a technologist's eye view). Note that the insane economic bubble has very little to do with the underlying technologies. LLMs are impressive but not very useful; other areas of AI/ML that don't get the gosh-wow coverage—pattern reconizers—are potentially revolutionary for a variety of fields.
-
RE: https://mastodon.nl/@bert_hubert/116000388750823868
A reasonable summary of the field (a technologist's eye view). Note that the insane economic bubble has very little to do with the underlying technologies. LLMs are impressive but not very useful; other areas of AI/ML that don't get the gosh-wow coverage—pattern reconizers—are potentially revolutionary for a variety of fields.
-
RE: https://mastodon.nl/@bert_hubert/116000388750823868
A reasonable summary of the field (a technologist's eye view). Note that the insane economic bubble has very little to do with the underlying technologies. LLMs are impressive but not very useful; other areas of AI/ML that don't get the gosh-wow coverage—pattern reconizers—are potentially revolutionary for a variety of fields.
@cstross We do have to spend such a lot of time fighting shit that the worthwhile machine learning gets thrown out with the bathwater.
tl;dr They can be useful for finding questions. Beware of their proffering magical answers.
-
@cstross I'm not really fond of him saying Whisper's performance is "super human" because it cleaned up a messed-up sentence in transcription.
-
@bert_hubert The specific example you attached that term to is not demonstrative of the term. But, if I wanted to bother, I could probably name you a *transcription firm* that did this. If you paid them. @cstross
-
@bert_hubert The specific example you attached that term to is not demonstrative of the term. But, if I wanted to bother, I could probably name you a *transcription firm* that did this. If you paid them. @cstross
@bert_hubert In any case, I refuse to play. If I robbed a bank, I could do whatever I wanted with those millions. I could pay dozens of transcribers. Doesn't mean that suddenly it's fine.
Unless of course Whisper AI is ethically sourced. Nevermind then.
-
RE: https://mastodon.nl/@bert_hubert/116000388750823868
A reasonable summary of the field (a technologist's eye view). Note that the insane economic bubble has very little to do with the underlying technologies. LLMs are impressive but not very useful; other areas of AI/ML that don't get the gosh-wow coverage—pattern reconizers—are potentially revolutionary for a variety of fields.
@cstross 99% of this in honesty, belies that Bert has significant expertise and knowledge in "AI." He does not. And it's not at all reasonable, because it makes simply false claims. LLMs are not impressive at all; we had ELIZA in the 80's. That's all they are.
Actual AI research has been set back decades, if not further, by the promotion of hacks and quacks.
Actual ML work has also been set back decades, as it gets tossed for 'cheaper' LLMs. -
@cstross 99% of this in honesty, belies that Bert has significant expertise and knowledge in "AI." He does not. And it's not at all reasonable, because it makes simply false claims. LLMs are not impressive at all; we had ELIZA in the 80's. That's all they are.
Actual AI research has been set back decades, if not further, by the promotion of hacks and quacks.
Actual ML work has also been set back decades, as it gets tossed for 'cheaper' LLMs.@rootwyrm Current LLMs go a lot further than eliza, although they're no closer to being intelligent. I doubt they're useful outside some very limited problem domains (eg. automated character interactions in MMOs).
-
@cstross 99% of this in honesty, belies that Bert has significant expertise and knowledge in "AI." He does not. And it's not at all reasonable, because it makes simply false claims. LLMs are not impressive at all; we had ELIZA in the 80's. That's all they are.
Actual AI research has been set back decades, if not further, by the promotion of hacks and quacks.
Actual ML work has also been set back decades, as it gets tossed for 'cheaper' LLMs.@cstross "but LLMs can" LLMs are a bad ELIZA with a big dataset. In the late 90's to 00's I experimented fairly successfully with something that took the day's news, chopped it up into tokens, and estimated the movement of a stock based on it. It's the same shit, except it ran on surplus 200MHz Pentium Pros and 10k RPM SCSI.
And I based that on shit from, god, mid 80's had to be. Maybe late 80's.And the shit these hacks have can't even do that.
-
@rootwyrm Current LLMs go a lot further than eliza, although they're no closer to being intelligent. I doubt they're useful outside some very limited problem domains (eg. automated character interactions in MMOs).
@cstross oh, hooboy, yeah no. You even think about trying to do that with the chatbots, and you'll have to flee your home from enraged gamers. Doubly so when these idiots are so bad at writing code they need two dozen GPUs and a terabyte of RAM just to give the wrong answer. And if you think gamers hate "AI" bullshit now, screw with their actual experience in-game.
Yeah. That won't go well. -
RE: https://mastodon.nl/@bert_hubert/116000388750823868
A reasonable summary of the field (a technologist's eye view). Note that the insane economic bubble has very little to do with the underlying technologies. LLMs are impressive but not very useful; other areas of AI/ML that don't get the gosh-wow coverage—pattern reconizers—are potentially revolutionary for a variety of fields.
@cstross Very amusing when two of my favourite people from radically different interests somehow end up interacting on Mastodon.
-
R ActivityRelay shared this topic
