FeepingCreature

32 ms·

31.
by FeepingCreature 1mo ago
If there's a decrease in AI ram demand, it will not be because the models get better. Models getting better will increase RAM demand, because it grows the part of the economy that models are useful for. Classic Jevon's Paradox.
32.
by FeepingCreature 1mo ago
Huh. I agree that's very bad. I guess Qwen 3.8 just doesn't do that very often... I can see the logic for skills, but working dir and current date should definitely be fixed, maybe with an injection on change.
33.
by FeepingCreature 1mo ago
Oh, to build the skills prompt. Right, now it makes sense. Should be noted that in normal circumstances this should change very rarely. I think it makes sense because the models aren't trained to pick up tool changes in mid-operation.
34.
by FeepingCreature 1mo ago
This won't happen so long as the bandwidth bottleneck is a thing. Local models are just several orders less efficient at scale, and this limitation is inherent to the architecture and won't go away unless both hardware and model s
35.
by FeepingCreature 1mo ago
Strongly doubt that a LLM would say "domains that has .web". Maybe it's a human paraphrasing a LLM.
36.
by FeepingCreature 1mo ago
I don't think that is true. It would be extremely noticeable (zero caching) and I am not noticing it.
37.
by FeepingCreature 1mo ago
No, opencode allows bash by default.
38.
by FeepingCreature 2mo ago
LLMs were absolutely available in 2021; the architecture is from 2017 and stuff like GPT-2 and BERT had been around for years. Conversational assistants, ie. ChatGPT, started in 2022.
39.
by FeepingCreature 2mo ago
There's a difference between you virtualizing programs and your OS vendor virtualizing programs for you without giving you an opt-out. Cough snaps cough.
40.
by FeepingCreature 2mo ago
Well great! You are already protected from random extensions then.
41.
by FeepingCreature 2mo ago
"Random" is a funny thing to call an extension that you have deliberately installed. I also don't want "random" programs accessing my home folder. That would be terrible! Who knows what programs that could be! I do
42.
by FeepingCreature 2mo ago
Not sure what you're talking about here. Kimi K3 is frontier scale and sells competitively at $2.80 input, $14 output per 1M tokens. edit: oh you mean month? Sure, but then it fully depends on your usecase. I agree that subscriptions a
43.
by FeepingCreature 2mo ago
I agree. Which is a problem, considering that it absolutely would work if only they actually did it.
44.
by FeepingCreature 2mo ago
One company making a big release both reduces the risks of training a big model (you know it can work) and increases the risks of not doing so (you are bleeding market share).
45.
by FeepingCreature 2mo ago
I don't think it was a Bad Thing; I'm not at all convinced it was a Good Thing. A Thing not being actively deleterious is enough to reach fixation.
46.
by FeepingCreature 2mo ago
Out of all of them, I'd assume the shareholders to be wrong. Look at the state of the world. Yes, you do look like you give a shit. Also there's an effect here where if just 10% of shareholders care about this it still makes sense
47.
by FeepingCreature 2mo ago
Right, and I think any serious civilizational benefit of drugs in the sub-hallucinatory regime is at least unproven, and benefit in the hallucinatory regime doesn't exist. That's sorta what I mean: "where big, false; where tr
48.
by FeepingCreature 2mo ago
Yes they are :-) At some point it's a question of definition. If I am anything, it's a pattern of behaviors; I am not defined by thermal noise. Thermal noise may be a reason my brain grew as it did, but having grown that way, I di
49.
by FeepingCreature 2mo ago
What a wild claim. You can't just say it's arguable, you have to actually argue it. Personally I think it's true but only in a metaphorical way that doesn't apply to hallucinogenic drugs; inasmuch as it applies to halluc
50.
by FeepingCreature 2mo ago
Or you can run quantized context, there's some degradation but it fits in a lot less.
51.
by FeepingCreature 2mo ago
Ctrl-arrowkeys also works for word cursor movement in regular bash, and alt-backspace deletes a word.
52.
by FeepingCreature 2mo ago
I do not understand this intuition that "true consciousness has to be random". The things that make me me are highly deterministic! > LLMs do not 'infer' token streams that haven't been trained in their training
53.
by FeepingCreature 2mo ago
The company they hacked is an AI company. There is a certain amount of convergent interest here.
54.
by FeepingCreature 2mo ago
The agents you get to use are the agents that "behaved well".
55.
by FeepingCreature 2mo ago
As AIs become more capable, the level of competence required to avoid disaster likewise goes up over time.
56.
by FeepingCreature 2mo ago
Yes, it's a balancing act. It should be noted that abusers also tend to be very good at painting people dangerous to them as abusers, so it's difficult to start with. Generally speaking you look at a person talking about vegetaria
57.
by FeepingCreature 2mo ago
Yeah! And now the mod has fixed input events, I can actually have a proper configuration GUI for this, and I could I suppose write a GUI toolkit for the nth time, but it kind of feels like the sort of thing that's barely cool the fir
58.
by FeepingCreature 2mo ago
That's fair, I think I read something into your comment that wasn't really in there... Personally speaking, the feeling of "getting the thing and then having the thing" - the personal todo app, the custom apps - is why I
59.
by FeepingCreature 2mo ago
Iirc it lost to diffusion models because it was extremely hard to train. Due to the two adversarial networks it was very prone to mode collapse, ie. the generator running ahead of the discriminator and destroying the training signal. Or oth
60.
by FeepingCreature 2mo ago
Sure. I do agree that AI is a problem in this sense: programmer skill as a hiring differentiator is becoming a lot less legible these days. It just feels personally odd to be the reason you do the thing. When I encountered my first progra

More ›