sigbottle

5 ms·

1.
by sigbottle 3d ago
To a first approximation, everything is power. How do we prove alignment? Often times, large regulatory bodies. Who supports that? Does the regulatory body actually have enough power to enforce those standards ? (Hint: look at modern Ameri
2.
by sigbottle 5d ago
For me, I've honestly never been much of an engineering builder - I'm focused on learning for myself. Which is useful - AI has accelerated that a ton - but it's still bottlenecked by, unfortunately, me. I should probably look
3.
by sigbottle 5d ago
May the iterative loop of adding new axes to evaluate on be a natural, healthy progression, instead of needing to frame it as an us-them problem? If you value humans intrinsically, this is necessarily the loop that will converge. I don&#x27
4.
by sigbottle 6d ago
I find the discourse under this blog fascinating. > The big, old ideas about intelligence that ended up basically vindicated were the ideas about how intelligence is about prediction, and prediction is about compression, and compression
5.
by sigbottle 6d ago
In general, a lot of moral invariants that natural selection has rendered as "intuitive" to us are no longer intuitive or possible. These natural brakes are not braking.
6.
by sigbottle 7d ago
At the end of the day, DSP is a lot of math and a lot of hardware. A lot of GNU Radio's code is a monster unification of so many different choices into certain interfaces (or duplication of the same choices) for, presumably performance
7.
by sigbottle 7d ago
Nit: is it any computable function? I thought the requirements were unbounded (in principle) memory and time. (For all intents and purposes given how high dimensional you are and using the "vibes" of computability yes I agree w&#x
8.
by sigbottle 7d ago
Chivalry and honor. Fight them to a duel to the death on horseback and javelin jousting.
9.
by sigbottle 8d ago
> "we demonstrate that LLMs can spontaneously develop novel social biases about artificial demographic groups even when no inherent differences exist" For a while (It's getting better with Astra, but still there), a lot of
10.
by sigbottle 8d ago
Well, according to Terry Tao, there were recent developments (from weeks ago) that made Navier Stokes in principle, solvable. So ignoring time, I say possibly, just because the groundwork was laid. What's impressive is parallelizing it
11.
by sigbottle 8d ago
It's specifically the last two bullet poitns - Tristan is suspicious of the timing, as only few others were trying this approach. OpenAI says the model didn't access his user data directly, but leaves unanswered whether Tristan&#x
12.
by sigbottle 12d ago
Chain of thought reasoning is really good, true, but agents really took off with tool calling and being able to integrate with a bunch of stable tools. In some sense, creating good new abstractions externally and learning those is a form of
13.
by sigbottle 12d ago
I think it goes to show how different notions of "compression" can be (that word gets thrown a lot nowadays because of AI). Humans like simplicity for sure, but there can be different use cases for the simplicity
14.
by sigbottle 13d ago
NVM I learned about leelaXodds today
15.
by sigbottle 13d ago
Oh wow, this is one of the best explanations on the thread. I guess the horizon problem is not such an issue for chess bots because reaching endgame is much shorter? And whole-piece handicaps probably snowball way faster than a potential go
16.
by sigbottle 13d ago
There's a DNS wizard at my job (not doing DNS stuff currently; but in his past life), and while he was talking to me about certain topics my eyes glazed over, and I thought, "Man, surely that won't affect me, right?" Wel
17.
by sigbottle 15d ago
I mean, aside from all the NP Hard sub problems (I might not be reading your satire though), the real hard problem is that a "product" is not just software, but an ecosystem and a promise of maintenance of the abstract user experi
18.
by sigbottle 15d ago
That's true. Again, I currently view LLMs as a function of integration - what they may lack in "intrinsic smarts", whatever that means, they can tool call and we build capacities (and they build capacities!) around them and t
19.
by sigbottle 15d ago
Kind of hijacking, would you say that LLM's have solved the frame problem? To me, the frame problem is: Can you function in an open vs closed world, and to me the answer is yes, LLM's can definitely function in an open world where
20.
by sigbottle 17d ago
Code (not AI) in general is the worst at semantics. Not an excuse for their practices, but it makes sense they wouldn't try to figure out something more advanced. If they did, their solution would probably be something like, "try
21.
by sigbottle 17d ago
> You can maybe automate getting something plausible out, but how would you know if it was _good_ in a timely way? Contrary to the other guy, the answer is accountability, either personal, or higher-level systems like competition and ult
22.
by sigbottle 19d ago
Yeah, philosophy of science & analytic philosophy (especially with Quine) has been down this rabbit hole before. Formalisms should be in service to higher-level intelligences, not the other way around. It's pretty clear that LLM&
23.
by sigbottle 19d ago
I mean, we ground that in external verification (or, use math, which in theory is accessible to everyone, but also has an "empirical" loop now via lean). I'm not saying AI has remotely solved this problem, not even close. But
24.
by sigbottle 19d ago
Oh, it is the shoelace formula, but for 3D? OP's description vaguely triggered memories of the shoelace formula from a decade ago, but geometry was never my strong suit. All I remembered was positive/negative triangles (or was it
25.
by sigbottle 27d ago
It still feels academic though. Suppose your response to that is, "Well, suck it up, its formalized". I'd argue that it'd be worse if you could teach your book, and a 50 year old who both has rich lived experience and
26.
by sigbottle 27d ago
I think software, paradoxically, has deeper network and intertial effects than pretty much anything else we have invented so far. Compare the leverage it enables to the ecosystem lock in it forces. With physical goods, it's not nearly
27.
by sigbottle 28d ago
Optimizing for the appearance rather than the underlying substance. Ahh. I know Terrence Tao didn't have the following in mind when he said those words, but oh man, the philosophy neurons are firing in my brain rn.
28.
by sigbottle 28d ago
More generally, in a system of agents, you could impose deep structure by construction on all the agents, or figure out from a set of black box agents, necessary convergence principles. Or you could treat them as black boxes, but instead of
29.
by sigbottle 1mo ago
Honestly, it kind of reminds me of my life. I need to start going to the gym, stop being a slob, start showing up to my job not like I'm a college freshman attending classes but as a respectable young man. But instead of doing that, I&
30.
by sigbottle 1mo ago
Well, GPU kernels are co-designed really hard. A lot of it is, async tile pipelines + spam my MMA primtives. Obviously it's still hard , but the point is that, by construction (cause like, NVIDIA literally releases primitives like thi

More ›