drdeca

51 ms·

661.
by drdeca 3y ago
I think it maybe needs to have been spoken with the intent to communicate or express a somewhat specific thing. Like, a couple seconds ago I vocalized the sounds “xzoubadoo”, but I didn’t attach any particular meaning to it, it was just an
662.
by drdeca 3y ago
Well... , yeah, but, these models already produce errors for other unrelated reasons, and like... Well, what exactly would we be showing that these models can’t do? Quines exist, so there’s no general principle preventing reflection in gene
663.
by drdeca 3y ago
While, uh, no , regardless, that has no bearing on my point, which is that your point claiming that [it is reasonable to treat A as evidence for B, but irrational/religious to treat B as evidence for A], is entirely wrong[1]. [1](.
664.
by drdeca 3y ago
To not work on anything where the risks from its potential weaponization outweigh the likely benefits of beneficial uses?
665.
by drdeca 3y ago
What leads you to suspect that Gödel incompleteness may be relevant here? There's no formal axiom system being dealt with here, afaict? Do you just generally mean "there may be some kind of self-reference, which may lead to some k
666.
by drdeca 3y ago
P(A|B) * P(B) = P(A & B) = P(B|A) * P(A) (Under the assumption that P(B) > 0 and P(A) > 0 ): P(A|B) > P(A) iff P(A|B) * P(B) > P(A) * P(B) and so iff P(A & B) > P(A) * P(B) and therefore, by the same reasoning in the
667.
by drdeca 3y ago
I am a Firefox user, and I think it better than the alternatives I'm familiar with, and I think I at least partially know what they are talking about? I think the "opens tabs all the time telling you how much Big Browser care abou
668.
by drdeca 3y ago
Do you mean just that it is very very likely that there is, or is there some argument that the MD5 function must be surjective?
669.
by drdeca 4y ago
Would the natural analogy of “tokens” be “words”, or something more like, “portion of mouth-movement”?
670.
by drdeca 4y ago
Even if it is all your own code, there's the joke: There are two parts to programming. The second part is debugging, where you remove the bugs. The first part is when you put the bugs in. I didn't tell it quite right...
671.
by drdeca 4y ago
> Rotational symmetries lead to the law of conservation of momentum. Conservation of angular momentum. Conservation of momentum corresponds to spatial translation symmetry, not rotational.
672.
by drdeca 4y ago
Banning instructions on how to make bombs would be censorship. Banning the making of bombs is not censorship.
673.
by drdeca 4y ago
Oh! This is applying FFT to the images! I thought it was going to be applying to functions f_n : [0,2pi] -> [0,1]^2 where f_n is the n-th level version of the curve. I guess in that view, higher order curves would have higher frequenci
674.
by drdeca 4y ago
I’ve seen it on signs on campus.
675.
by drdeca 4y ago
What’s reversed about it? I guess the fact that it is a reason to not introduce a new thing rather than a reason to not-get-rid-of a long-kept thing?
676.
by drdeca 4y ago
I thought it was going to be about implementing transformer models in a way that fits well with some special kind of hardware like those green arrays or whatever they are called (the low energy use, highly parallel (but maybe low clock spee
677.
by drdeca 4y ago
Consider the case where the atom in question is hydrogen, with just one electron orbiting it. I would still think the word "cloud" would be used?
678.
by drdeca 4y ago
The article refers to a gravitational something tensor, and, I’m wondering whether that is because this is actually dealing with gravitational effects, or whether it is just called that because it is the same tensor (or basically the same,
679.
by drdeca 4y ago
Following the safety techniques should be helpful to the goals of most malicious actors, provided that they prefer being in control over an alien optimizer being in control? Granted, if the safety/alignment techniques have too large of
680.
by drdeca 4y ago
Bad humans taking over the world is still better than some inhuman alien optimization process taking over the world.
681.
by drdeca 4y ago
I think I generally understand the transformer architecture. Now, "developing their own goals", maybe that wouldn't make sense for LLMs alone, but "planning how to achieve [some goal]", seems somewhere between &quot
682.
by drdeca 4y ago
If the safety groups figure out how to do things in a way which we can be confident is safe, this would make it at least possible for the capabilities researchers to do things in those ways. I would imagine people would prefer to do things
683.
by drdeca 4y ago
I think you are talking about enforcing different things? The thing to enforce, which the parent commenter described, seemed to be a 20% tax on AI profits of “big AI companies”, to be paid to AI safety research groups. That doesn’t sound un
684.
by drdeca 4y ago
Do you have a story about how self-driving cars could lead to an x-risk?
685.
by drdeca 4y ago
I just meant some correlation, not necessarily a particularly strong one. I didn’t mean to suggest that there are no people who are both highly intelligent and also completely vile.
686.
by drdeca 4y ago
My impression is that, among humans, there is a somewhat positive correlation between intelligence and benevolent behavior. (But I would expect the tails to come apart somewhat. Also, among non-human minds I wouldn’t necessarily expect thi
687.
by drdeca 4y ago
Consumers of the API
688.
by drdeca 4y ago
Yes, I do think that is to get more training data... But my impression is that continual training is difficult, so it makes more sense to get a big chunk of it and then do a big chunk of training on that? But maybe they have done that at le
689.
by drdeca 4y ago
I don’t think it is being actively trained on the current conversations? Or at least not in a continual way...?
690.
by drdeca 4y ago
Huh; I had assumed it would refuse to do that?

More ›