hyperpape

6 ms·

31.
by hyperpape 2mo ago
> Huggingface covered it up They announced it publicly within days. https://huggingface.co/blog/security-incident-july-2026
32.
by hyperpape 2mo ago
Did you describe what happened in a neutral tone?
33.
by hyperpape 2mo ago
A Modest Proposal that will never be implemented: add an interface to String, Boolean, Integer, Long, Float and Double. Because all these types already implement "toString" (and I believe their toString representations are compati
34.
by hyperpape 3mo ago
I'd slightly disagree. Here's an example of a programming task that is incredibly complicated: Your job is to maintain and modernize a system while delivering a constant stream of new features. Your system is several million lines
35.
by hyperpape 3mo ago
I studied math through college before learning to program as an adult and becoming a software engineer, and I strongly disagree. I don't know how to say this in a way that won't sound insulting, but I don't mean it to be insu
36.
by hyperpape 3mo ago
The attacking models don't have access to all the data that OpenAI has. Like, they don't say "hey Sol, here's the password to SamA's bank account."
37.
by hyperpape 3mo ago
> It's early in the year. You want to drive straight through the middle of Chinatown in SF. Why might that be a bad idea? The premise seems to be that models aren't smart enough to understand this, and if they were, they'd
38.
by hyperpape 3mo ago
Different people should use different words. I don't think a content creator would naturally describe themselves as a content creator. The word is useful for contexts where you need to generalize over all the different types, but that&
39.
by hyperpape 3mo ago
> $168k Billions of dollars of annual revenue go through Claude Code, and the people who work on it must be a lot of millions in headcount. The time matters a fair bit, it's probably time someone can't spend breaking things in
40.
by hyperpape 3mo ago
Related, but not quite the same. The solution to this problem can have more than one loop, whereas TSP definitionally has only one. Not quite sure how that changes the difficulty or the tools that you’d use.
41.
by hyperpape 3mo ago
> And if that's the case, the LogEvent objects will end up automatically promoted to OldGen. Why do you think this would happen? There's no mechanism that makes young gen objects that reference old gen objects (or are reference
42.
by hyperpape 3mo ago
1. A lot more than 1000, you're off by more than one order of magnitude. It's definitely beyond my level of graph theory knowledge (undergrad level) but looking at the paper, it's not using any crazy machinery, and it's
43.
by hyperpape 3mo ago
> rejected any notion of utility. It would be fundamentally wrong for you to ask what's the value of solving the Erdős–Hajnal conjecture; the value is that it's solved. I disagree. Mathematicians care about the utility of a res
44.
by hyperpape 3mo ago
Personal and professional are not mutually exclusive. If I criticize your code, that is a professional criticism. If I criticize your code and say it reflects your consistent carelessness and stupidity, it is also personal. If I say you fab
45.
by hyperpape 3mo ago
Limiting the options an LLM has does not turn it into a menu, because it can create infinite combinations/chains of behavior based on the items that it has. Of course, that power also makes it harder to anticipate security issues--if y
46.
by hyperpape 3mo ago
Maybe there's an argument that a lack of rising profit margins in non-tech companies is a bad sign for AI, but this article doesn't make it. Why can't we have a red-queen's race where non-tech companies are implementing
47.
by hyperpape 3mo ago
Although it's not the most important thing here, I love the IQ badge on the right side of the screen. At IQ 200, the AI is finally qualified to carry a pager for production. I'm in the pagerduty lineup, and shockingly, my IQ isn&#
48.
by hyperpape 3mo ago
Sounds like the real number is 454x faster, not 70x. Checkmate, nerd!
49.
by hyperpape 4mo ago
True, but it's not relevant because that isn't how we actually train LLMs for use as quasi-intelligent tools. We specifically do not want the model to be able to just memorize its input, which is what your process requires. Many t
50.
by hyperpape 4mo ago
> A shift is happening among major AI labs, who are becoming increasingly skeptical of endless parameter count and training data scaling. The limits of this paradigm were put on the world’s stage when Claude Fable 5 was restricted by the
51.
by hyperpape 4mo ago
I think this is an analogy that's been taken far too far. The output of intelligence just isn't compression, that's memorization. The role of intelligence is to generate novelty. It's true that LLMs do something that loo
52.
by hyperpape 4mo ago
I’m a hallucination. None of these are me. Perhaps the closest is DeepSeek v4: > Hyperpape is a user on the LessWrong forum, known for thoughtful comments on rationality and philosophy. I studied philosophy, so maybe, except I don't
53.
by hyperpape 4mo ago
This article is inane. The law seems...truthy, though I find it a little too underspecified to assess. The conclusion is true, and I'll even overlook it being a slight strawman. Your "long genius"[0] CEO cannot talk to AI and
54.
by hyperpape 4mo ago
First sentence: > In my Ottawa life, every Tuesday evening, I take two gym classes back to back—boxing and the pompously named “body sculpt,” which makes me discover muscles I didn’t know I had. The em-dash matches how you'd speak o
55.
by hyperpape 4mo ago
The Mac is a pre-existing platform that is both more capable than iOS, and had an existing user base that used apps that had much greater access. Apple’s attempts to lock down the Mac have met with poor adoption. In exchange, it also less s
56.
by hyperpape 4mo ago
> The curve of AI improvement will continue at the current pace I guess this is trivially true if you say "maximalism" (hell, the maximalists think it will speed up as the AI becomes a super-AI-researcher), but as long as the r
57.
by hyperpape 4mo ago
We need a companion to "IN MICE", which is "IN EVALS". I don't think this is bad research, but you have to understand how far it generalizes. I'm not saying that evals are useless, we need to do our best to pro
58.
by hyperpape 4mo ago
> According to the paper, “Of the 22 vulnerabilities, five were level-based, meaning that the default weak isolation level led to the anomalies behind the vulnerabilities. The remaining 17 were scope-based, meaning that the database acce
59.
by hyperpape 4mo ago
I wonder what the impact of the rising base rate of employees with college degrees is. In 1992, a fresh college graduate had better educational attainment than 42% of the labor force. In 2016 (latest date I found numbers for), that was down
60.
by hyperpape 4mo ago
> If AI tokens were so magical in creating new value in developing software applications generally, they wouldn't be selling tokens directly. They'd hoard the tokens are use them to dominate SaaS software in any industry they w

More ›