aesthesia

4 ms·

1.
by aesthesia 5d ago
I don't get this take. They have something horribly dangerous, but regulating it might benefit them in some way, so therefore we should do nothing?
2.
by aesthesia 5d ago
I don't see how this is relevant. OpenAI absolutely had sloppy security practices here. But that doesn't mean there was no "real security", and it certainly doesn't mean that their models were simply following order
3.
by aesthesia 5d ago
> Do you accept that the story / justification from the labs in the popular media and political discussion is simply nonsense? No, and I don't see anyone who's actually demonstrated understanding of what happened in the Hu
4.
by aesthesia 5d ago
I would rather not turn the internet (or the rest of existence) into a dark forest if we can help it. Are you sure that's not preventable?
5.
by aesthesia 5d ago
More than one thing can be true. OpenAI was absolutely negligent, but this was only able to happen because the models were capable and persistent, and had a tendency to go far beyond any reasonable boundaries. And, importantly, OpenAI'
6.
by aesthesia 5d ago
The failure mode was _not_ identical. The HF incident agents were not directly connected to the internet and had to compromise an internal package registry in order to access the internet.
7.
by aesthesia 8d ago
From what I can tell these are classifiers trained for a single task. What excites people about Jev is that it can do zero-shot structured responses for arbitrary prompts. Now, this isn't new either; models like GLiNER have been around
8.
by aesthesia 9d ago
The question is whether the goal is to stop (or delay) China from developing its own AI chips or to stop (or delay) Chinese AI development in general.
9.
by aesthesia 9d ago
Sort of? There's a blurb here with a link to a tweet: https://openai.com/hugging-face-incident-and-misalignment/#m...
10.
by aesthesia 9d ago
Their explanation of this behavior is pretty interesting, actually. ( https://alignment.openai.com/misalignment-reports/self-gener... ) > The cases clustered around a few training steps and coincided with a spike in “
11.
by aesthesia 9d ago
Wouldn't you then need to solve the similar problem of where to store the brine in the meantime?
12.
by aesthesia 9d ago
This isn't a political article?
13.
Encouraging Deception in Compaction Summaries(alignment.openai.com)
2 points by aesthesia 9d ago|0 comments
14.
by aesthesia 9d ago
A big part of this is the hyperstition argument: discussion about different properties AI models could have in the training data may become a self-fulfilling prophecy. There are similar concerns about discussions of AI misalignment in train
15.
by aesthesia 10d ago
Well, what you get when you don't specify anything in training about how models should respond to questions like this is LaMDA: https://en.wikipedia.org/wiki/LaMDA#Sentience_claims . Every training method is puttin
16.
by aesthesia 10d ago
So, like ARC-AGI-3? I'm sympathetic to the notion that the things we're able to measure are necessarily going to miss important aspects of capability and intelligence. But people are attempting to measure this kind of thing, and m
17.
by aesthesia 10d ago
I'm also not sure that we know a non-number can never be Numberwang.
18.
by aesthesia 11d ago
There's very little reason to believe Irregular was involved in that incident, and the article presents no evidence to that effect. So it's "There's not a teapot orbiting the sun somewhere between Earth and Mars (as far
19.
by aesthesia 11d ago
Oh, the headline is technically correct: there was an OpenAI incident that Irregular was involved with, disclosed shortly before the Hugging Face one.
20.
by aesthesia 11d ago
OK, I still have no idea what you actually think the "correct" answer is.
21.
by aesthesia 11d ago
One thing glossed over in this article is that Irregular was not involved in the OpenAI–Hugging Face incident; this seems like important context to share.
22.
by aesthesia 11d ago
This feels like an xkcd 169 situation, to be honest.
23.
by aesthesia 12d ago
> The profitability of being able to serve a query wasn't really under question In one sense, yes, but I do see people question it regularly.
24.
by aesthesia 14d ago
...by the version number? How are Claude Opus 4.8 and Claude Opus 5 differentiated?
25.
by aesthesia 15d ago
Ah, I see now that you were making a claim narrowly scoped to DeepSeek models specifically. Still, Anthropic has made specific claims about deliberate access to Claude CoT by DeepSeek (e.g. https://www.anthropic.com/threat-i
26.
by aesthesia 16d ago
I don't see how your links support the claim that the studied models did no distillation from US models.
27.
by aesthesia 16d ago
This is a clear dig at OpenAI: > We have signed an agreement with METR to conduct an independent investigation of these incidents. Our agreement grants METR wide-ranging access, including to transcripts beyond the window in which the inc
28.
by aesthesia 17d ago
What's the fundamental constraint that will ensure this continues to be the case in a year, or five years?
29.
by aesthesia 17d ago
@hilbertspaess is not a nonsensical user name. The accounts he follows are totally reasonable for an AI researcher. I think it's extremely believable that he created an account in January, followed a few people as part of the initial s
30.
by aesthesia 17d ago
> In the space of about four years, LLMs progressed from mediocre high school student to Ph.D. graduate in every field. That's impressive, but there is no evidence yet they can outperform or outsmart humans. I mean, unless you see c

More ›