Post AhxAfpuG5gRorbmiPI by remenca@mastodont.cat
(DIR) More posts by remenca@mastodont.cat
(DIR) Post #Ahw1Lpo1f40k0yIUBk by Wolven@ourislandgeorgia.net
2024-05-16T02:31:12Z
0 likes, 2 repeats
OpenAI has been outright lying to everyone about their capabilities and accomplishments, which: We told you.But I do appreciate @Julia always doing the solid work to lay it out plain.
(DIR) Post #Ahw3KRYGYINcPVOF2u by ben@m.benui.ca
2024-05-16T02:53:20Z
0 likes, 0 repeats
@Wolven @Julia something half-decent published in the NYT? I guess a broken clock is right twice a day
(DIR) Post #AhwGDGba9S6ppyqimm by macallik@federation.network
2024-05-16T04:55:38.330Z
0 likes, 0 repeats
@Wolven@ourislandgeorgia.net @Julia@journa.host Does that changes how you view gpt 4o?
(DIR) Post #AhwGDIwvR4N16kqnKa by Wolven@ourislandgeorgia.net
2024-05-16T05:17:45Z
0 likes, 0 repeats
@macallik @Julia No, because I already thought it was trash
(DIR) Post #AhwxAxdZzxLwSYvkpM by remenca@mastodont.cat
2024-05-16T13:19:05Z
0 likes, 0 repeats
@Wolven @Julia Without dismissing the fact that google and openai are the worst, I do not understand how having a machine ranking in the 48th percentile, this mean matching the average human, is not perceived as a feat of the highest degree.
(DIR) Post #AhxAfpAAr1aIYgDuZk by senil888@furry.engineer
2024-05-16T14:19:52Z
0 likes, 0 repeats
@remenca @Wolven @Julia It's more about the lying part - "acing" the bar exam makes people think it must've done really really good, but in reality it just did as well as anyone else by more or less guessing off the question. And I imagine most people not actively preparing for the bar would be doing that. Informative guesses, but guessing nonetheless. "It managed to do as good as George over there who didn't know shit and guessed as good as he could" doesn't really sell it as some magical replacement tech that the hype claims.Also, the thing probably has a ton of legal stuff it could allegedly reference if poked right, it could just know the bar enough through that and it Worked Out Well. Because it had that knowledge anyways. Which most people, during the bar, won't have access too.In that view, you could argue it "cheated" the bar, and in doing so it matched what most people would do with probably limited legal experience. If you could remove that knowledge, it'd probably do worse. Maybe a lot worse.
(DIR) Post #AhxAfpuG5gRorbmiPI by remenca@mastodont.cat
2024-05-16T14:44:05Z
0 likes, 0 repeats
@senil888 @Wolven @Julia I think that this is precisely what I meant, like, 10 years ago the best AI systems in the world where suffering to distinguish a cow from a plane, and suddenly 8 years after we have machines that score the same as George. I mean, George might be ignorant, but sure know how a cow and a plane looks
(DIR) Post #AhxAfrq4uCIArDqPAW by Wolven@ourislandgeorgia.net
2024-05-16T15:50:23Z
0 likes, 0 repeats
@remenca Okay, so why not run with that? Why lie about it? And honestly "does as well as well as an automated coin flipper" just doesn't impress me. "Does as well as if lexisnexis flipped a coin" is a *bit* more impressive, but more for the fact of combining that collection of information with a statistical stochastic pattern matching system. Like… this could all be kind of cool if they didn't over-hype it and call it something it wasn't and then try to shove it into literally every aspect of human life at the cost of exploited human labour and massive environmental damage. @senil888 @Julia
(DIR) Post #AhxWMTo6EHKNoU4yBc by hybridhavoc@darkfriend.social
2024-05-16T19:53:22Z
0 likes, 0 repeats
@Wolven @Julia Saw a thread about this on Bluesky yesterday.https://bsky.app/profile/nafnlaus.bsky.social/post/3kskpqut6kf2z
(DIR) Post #AhxaB8GQNxpZd6AlJw by Wolven@ourislandgeorgia.net
2024-05-16T20:36:10Z
0 likes, 0 repeats
@hybridhavoc As I said elsewhere, '"Does as well as if lexisnexis flipped a coin" is a *bit* impressive, but more for the fact of combining that collection of information with a statistical stochastic pattern matching system.'Like… this could all be kind of cool if they didn't over-hype it and call it something it wasn't and then try to shove it into literally every aspect of human life at the cost of exploited human labour and massive environmental damage.' @Julia
(DIR) Post #AhxfG8pSDCEebIlVMe by remenca@mastodont.cat
2024-05-16T21:33:06Z
0 likes, 0 repeats
@Wolven @senil888 @Julia There are many interesting points to discuss here.- I fully agree with you that the blame is with the AI companies who hype the AI bubble. AI in its present state is already useful, no need to exaggerate it.- I'm not sure that you understand the concept of percentile. In short it means that if you order all the human contestants the AI will end in the middle of them. This is much better than flipping a coin, which probably would be in the bottom.
(DIR) Post #AhxgHK1cAcNLZJa1dQ by remenca@mastodont.cat
2024-05-16T21:35:56Z
0 likes, 0 repeats
@Wolven @senil888 @Julia - I never understood why the stochastic nature of machine learning is taken as something bad, as if human learning wouldn't involve stochasticity. We learn by trial and error in the same way of a machine. I mean, we use school book exercises to train AI. It is the same.- Again, we agree that capitalism is to blame, and that predatory practices like shoving AI everywhere to violate everyone's privacy is terrible. Trust me, I'm very mad because of this.
(DIR) Post #AhxgHLhpv7k8mYLG5I by Wolven@ourislandgeorgia.net
2024-05-16T21:44:32Z
0 likes, 0 repeats
@remenca LexisNexis being a repository of legal scholarship by people who, necessarily, passed the exam. Anyway, stop riding for large corps misrepresenting their performance, you're only shoring up the excuses they'll use to justify their next hype cycle. Bye.@senil888 @Julia
(DIR) Post #Ahxhqhzb9SR7zZ4fmC by remenca@mastodont.cat
2024-05-16T22:02:09Z
0 likes, 0 repeats
@Wolven @senil888 @Julia That's rich, stop YOUR excuses, blaming AI for what is capitalism fault.
(DIR) Post #Ahy1e4VG90fr7t8Rto by breadbin@bitbang.social
2024-05-17T01:43:55Z
0 likes, 0 repeats
@Wolven @Julia Passing a test when your source is the source material seems like something it should ace, now shouldn’t it?I don’t think ML/MMLs/etc can’t do anything, they clearly can. I just feel, as you say, it’s hype. Reminds me of Boeing, like the MBAs are in charge.Makes me sad if the search for profit means we are missing out on possible better use for these type of technologies. Things that could help people rather than reduce headcount.
(DIR) Post #Ahy41BTsZE6nKtddYm by keydelk@fosstodon.org
2024-05-17T02:10:32Z
0 likes, 0 repeats
@Wolven @Julia very interesting article. I’m also of the opinion that the technology would be rather interesting from a technical perspective (look at all the neat things you can do with fancy data analytics) if it weren't surrounded by so much hype and bullshit.