simonh

5 ms·

31.
by simonh 2mo ago
In the UK we've been lucky with our royalty recently, but with a different roll of the genetic dice we'd have Randy Andy as king instead of Charles. Anyway, you have kings, you just elect them and call the presidents, but they hav
32.
by simonh 2mo ago
Four Tribes is a decent iOS clone.
33.
by simonh 2mo ago
So a rounding error of nerds, who themselves are a rounding error, will invest a lot of effort and trouble to use tech that makes their and their user's lives harder, and probably won't even work, to hide text that LLM big tech co
34.
by simonh 2mo ago
>If it's functionally random PRNG, then how does it differ from any other random sampling? For practical purposes it isn't.
35.
by simonh 2mo ago
Going after every possible case would be staggeringly expensive for marginal gain, the cost society would massively exceed the benefits. Yes the system as it is that relies on discretion, but there’s always discretion involved, and we rely
36.
by simonh 2mo ago
But as I explained, the watermark is functionally random PRNG noise overlaid on the token probabilities. It’s not something that can be compensated for because it’s not predictable if you don’t have the seed and PRNG function.
37.
by simonh 2mo ago
It's Boy Genius Report, which used to be decent but was sold to a media conglomerate, with predictable results as you can see.
38.
by simonh 2mo ago
The Amiga was hardly a low end computer. I bought a 500 in 1989 and they were wonderful machines. The high end systems could run Unix and were used in commercial video production. You just can't compete with platform effects though, an
39.
by simonh 2mo ago
All those methods are applied with the specific goal of improving output quality and are applied to the extent that they do this. Watermarking has no such goal, and is not implemented for any such reason. In fact it's much more like ap
40.
by simonh 2mo ago
>You just need to know your LLM distribution and the encryption key, then with each new token you exponentially increase the chance of knowing whether it fits your encryption key. Without actually affecting the token choice in a perceiva
41.
by simonh 2mo ago
>It doesn't, though. From the outside, without access to the parameters, you can't distinguish the watermarking system from a random number generator. You can't as a user tell by how much the quality of the output was degr
42.
by simonh 2mo ago
>No one uses a pure random function over the whole probability distribution described by the LLM's output. So what? By definition with this system the LLM will chose tokens it otherwise would not, purely for watermarking reasons. Ye
43.
by simonh 2mo ago
The quality of the LLM just _is_ the quality of the token probabilities it generates. Better quality token probabilities, better quality output. Worse quality token probabilities, worse quality output. Watermarking changes the probability c
44.
by simonh 2mo ago
That cannot be true. The quality of an LLM's output is the quality of the probability calculations for the next token. Anything that degrades the relationship between the system's best assessment of the appropriate probability and
45.
by simonh 2mo ago
People saying AI are tools are not saying they are hammers. Obviously they act towards goals, but how and why they do so isn’t the same as for human psychological motivations. The moment someone interprets AI behaviour in terms of human psy
46.
by simonh 2mo ago
It’s effective in some ways, but can be very misleading in others in the ways I and others are discussing.
47.
by simonh 2mo ago
Both are true, even if they were aligned they still wouldn’t be doing what they do for reasons analogous to why people would. You’re probably right that we can’t fully understand how they function, and that will get harder, but it’s possibl
48.
by simonh 2mo ago
Do you genuinely think the AI is internally reflecting on its experience of “flinching” and reporting on a reaction it actually has? I don’t see any reason to believe this. Suppose you asked it to answer as though a character in a story had
49.
by simonh 2mo ago
There are two sides to this, there is the external behaviour and there is the internal process resulting in that behaviour. The internal process is not analogous to what happens in a person's mind when a person lies, and reasoning abou
50.
by simonh 2mo ago
What matters is that we accurately understand what these things are doing and why, otherwise we will keep on making mistakes both in how we build and train them, and in how we use them. In a sense you are right, it doesn't matter wheth
51.
by simonh 2mo ago
Does it's training token stream contain texts which talk about such things?
52.
by simonh 2mo ago
It's having a conceptual model of the world and of relationships between concepts beyond just relationships between tokens. I think OP explained this well, what does it mean if an LLM can recite the rules of a game verbatim, but cannot
53.
by simonh 2mo ago
To back this up, there is growing evidence that subsidised child care, financial support for families, etc has a minimal effect on birth rates, if any. https://link.springer.com/article/10.1007/s11150-026-09861-1
54.
by simonh 2mo ago
Thanks for the recommendation.
55.
by simonh 2mo ago
You got me there.
56.
by simonh 2mo ago
>The USA has never been defeated. It's never been taken over by force and forced to submit to a foreign power. War getting a little too expensive or politically unpopular for their tastes is not really a loss, and it implies nothing
57.
by simonh 2mo ago
Those are interesting points. The US military is optimised for limited strike "special military operations", which it does all the time, and deterrence. Neither of those require a super deep arsenal, just enough to complete a limi
58.
by simonh 2mo ago
The health care issue in the USA isn't due to lack of funding, the US spends more on health care per person than any other country, double that of many European countries with single payer systems. https://en.wikipedia.org&#
59.
by simonh 2mo ago
That's why I like the PBTA approach. All the rules you need are right there on the character sheet and a couple of pages of quick references sheets. Even with it's stripped down system, we spent plenty of time rulebook diving in D
60.
by simonh 2mo ago
I just picked up the books, after finishing playing in a short campaign. It's a great setting. The rules system is a bit bare bones, but you could easily run Dungeon World or pretty much any trad fantasy RPG there. Yes, 5e, sure. It wo

More ›