hexaga

13 ms·

31.
by hexaga 8mo ago
Being able to force someone to do something is not justification for doing so. Further, it is ridiculous to try and label that as 'beneficial for everyone involved'. By the same token you can call outright slavery under threat of
32.
by hexaga 8mo ago
Again, vacuous. You deride as 'metaphysical' what is psychological. But the health and well-being of children too is a 'metaphysical' concern to the worker by this metric, and yet you call it up to support yourself? Your
33.
by hexaga 8mo ago
A finely tuned set of heuristic triggers for fear, horror, disgust, etc. You might as well ask why pain is so painful.
34.
by hexaga 8mo ago
If I was in that position, and you gave me the choice to ritualistically mutilate myself for your amusement so my children could escape, I'd probably take it. Your entire chain of argument is vacuous; devoid of any sense of empathy for
35.
by hexaga 9mo ago
Because it's not calibrated to. In LLMs, next token probabilities are calibrated: the training loss drives it to be accurate. Likewise in typical classification models for images or w/e else. It's not beyond possibility to tr
36.
by hexaga 9mo ago
It's not the searching that's infeasible. Efficient algorithms for massive scale full text search are available. The infeasibility is searching for the (unknown) set of translations that the LLM would put that data through. Even i
37.
by hexaga 9mo ago
No person is forced, because a person's agency does not solely consist of the gap. It doesn't matter. The argument isn't: 'advertising is bad because it forces some specific person to do a thing they don't value&#x2
38.
by hexaga 9mo ago
Because advertising works. Full stop. It doesn't matter if it is valuable or not. It just works. Definitely not with P(buy this crap) = 1. But the effect is still there and real and measurable and google has made colossal amounts of mo
39.
by hexaga 9mo ago
Kokoro is fine tunable? Speaking as someone who went down the rabbit hole... it's really not. There's no (as of last time I checked) training code available so you need to reverse engineer everything. Beyond that the model is not
40.
by hexaga 9mo ago
I would like to make really clear the distinction between expressing an opinion and holding/forming an opinion, because lots of people in this comment section are not making it and confusing the two. Essentially, my position is that la
41.
by hexaga 9mo ago
I'd push back and say LLMs do form opinions (in the sense of a persistent belief-type-object that is maintained over time) in-context, but that they are generally unskilled at managing them. The easy example is when LLMs are wrong abou
42.
by hexaga 9mo ago
It's an efficient point in solution space for the human reward model. Language does things to people. It has side effects. What are the side effects of "it's not x, it's y"? Imagine it as an opcode on some abstract
43.
by hexaga 10mo ago
You've confused yourself. Those problems are not fundamental to next token prediction, they are fundamental to reconstruction losses on large general text corpora. That is to say, they are equally likely if you don't do next token
44.
by hexaga 10mo ago
If enough care about this that can and will do something about it (making formalization easier for the average author), that happens over time. Today there's a gap, and in the figurative tomorrow, said gap shrinks. Who knows what the f
45.
by hexaga 10mo ago
Learn what? I don't agree and you haven't given reasons. I don't write for your personal satisfaction.
46.
by hexaga 10mo ago
Alternatively: some people are just better at / more comfortable thinking in auditory mode than visual mode & vice versa. In principle I don't see why they should have different amounts of thought. That'd be bounded by ho
47.
by hexaga 10mo ago
Unavoidable: expecting someone else to do the connection isn't a viable strategy in semi-adversarial conditions so it has to be bound into the local context, which costs clarity: - Escaping death doesn't become more tractable beca
48.
by hexaga 10mo ago
I don't think it matters, to be quite honest. Absolute tractability isn't relevant to what the analogy illustrates (that reality doesn't bend to whims). Consider: - Locating water doesn't become more tractable because yo
49.
by hexaga 10mo ago
If wishes were fishes, as they say. To demonstrate with another example: "Gee, dying sucks. It's 2025, have you considered just living forever?" To this, one might attempt to justify: "Isn't it sufficient that dying
50.
by hexaga 10mo ago
Do selection dynamics require awareness of incentives? I would think that the incentives merely have to exist, not be known. On HN, that might be as simple as display sort order -- highly engaging comments bubble up to the top, and being at
51.
by hexaga 10mo ago
It just overlaid a typical ATX pattern across the motherboard-like parts of the image, even if that's not really what the image is showing. I don't think it's worthwhile to consider this a 'local recognition failure&#x27
52.
by hexaga 10mo ago
>> But we don't usually believe in both the bullshit and in the fact the the BS is actually BS. > I can't parse what you mean by this. The point is that humans care about the state of a distributed shared world model and
53.
by hexaga 10mo ago
The concrete itself can be damaged further over time by expanding root networks / growth.
54.
by hexaga 11mo ago
> if the body is experiencing adverse reactions to certain foods, the cause should be a biochemical one No. I can look at a picture of gross food, or imagine it, and be nauseated. Restricting potential causes this tightly is wrong. UPFs
55.
by hexaga 11mo ago
127.0.0.2
56.
by hexaga 11mo ago
There are different varieties of attention, which just amounts to some kind of learned mixing function between tokens in a sequence. For an input of length N (tokens), the standard kind of attention requires N squared operations (hence, qua
57.
by hexaga 1y ago
In the spirit of the library, which contains both your comment and mine: > The hypothetical "library of all possible books" isn't useful to anyone. That's not an archive, and has no uses even for researchers, especial
58.
by hexaga 1y ago
Why wouldn't it? That's downright in-distribution. Plenty of it in the pretrain corpus.
59.
by hexaga 1y ago
It doesn't matter / is not relevant. The harm is not caused by intent, but by action. Sending language at human beings in a way they can read has side effects. It doesn't matter if the language was generated by stochastic pro
60.
by hexaga 1y ago
Model weights are significantly larger than cache in almost all cases. Even an 8B parameter model is ~16G in half precision. The caches are not large enough to actually cache that. Every weight has to be touched for every forward pass, mean

More ›