FeepingCreature

7 ms·

1.
by FeepingCreature 5d ago
Well, it's rather more "if you're not a biologist, how do you know it can exist?" Things that don't exist yet are often very predictable. However, it simply seems to me that "Viruses can not condition on enviro
2.
by FeepingCreature 5d ago
I mean I don't know, I'm not a biologist. But existing viruses are not optimized for sneaky mass deaths.
3.
by FeepingCreature 8d ago
LLMs can learn from one sample.
4.
by FeepingCreature 8d ago
Yep, you'd want to put it on a timer. Optimally trigger it off some natural event, maybe temperature, so it can achieve universal penetration before activating.
5.
by FeepingCreature 8d ago
It seems like there would be a massive attacker bias in multiple ways. Defenders need consent, attacker does not. Defenders have to work with the human body, attackers only have to break it. Defenders have to stick to the law, which may pre
6.
by FeepingCreature 8d ago
"If unlikely thing happens, then an unrelated miracle will also happen" is not actually how logic works.
7.
by FeepingCreature 9d ago
Best I can do is all-powerful AI that's not even slightly aligned with anybody's values, sorry.
8.
by FeepingCreature 11d ago
Why not the other way around? First make sure the police department is capable of using them legally and reliably investigating abuse, then we can consider allowing them.
9.
by FeepingCreature 11d ago
> we need to stop giving agencies money based on stats. I think you just need to stop giving agencies money based on bad stats. Goodhart's Law is an extremely well-established effect. At the very least, crime rate should be monitore
10.
by FeepingCreature 12d ago
Or at least use a modern llama.cpp with KV activation rotation.
11.
by FeepingCreature 13d ago
Yeah, don't forget how dangerous GPT-2 was supposed to be. Able to generate realistic spam at arbitrary volume. You know, the thing that was 100% correct and actually occurred.
12.
by FeepingCreature 13d ago
I have a self hosted Qwen 3.8 27B and I find it unbelievably cracked and dedicated. It's at least credibly attempted everything I've thrown at it. Just today I had it write a toy compiler with a JIT backend just to test out a co
13.
by FeepingCreature 15d ago
Of course you can make inferences what the model is doing. The summaries are usually sufficient. They're summaries, not random noise.
14.
by FeepingCreature 15d ago
Yeah I think when it talks like this it's signaling to some (imagined) automated grader that it fulfilled a given constraint.
15.
by FeepingCreature 21d ago
Well, for one you're not supposed to use a majority vote to abolish democracy, or mess with unalienable rights. You can think of it as "democracy with precommitment".
16.
by FeepingCreature 21d ago
I like "and on the third hand." Like, yes I have three hands, what of it.
17.
by FeepingCreature 21d ago
The important thing about HuggingFace isn't the safetensors files, it's the READMEs and the Google search placement. It's easy to reproduce their distribution with torrents, it's very hard to then establish yourself as t
18.
by FeepingCreature 21d ago
People seriously undervalue the importance of "usefulness" in ontology. If I assume that the thing that others report is the same thing that I perceive, which given that they have a very similar slab of meat inside their skull as
19.
by FeepingCreature 21d ago
Any argument that would lead us to discard 99.9999999% of the phenomenon under debate, no matter how reasonable, is certainly not useful. That's why I hold with the philosophical stance of "consciousness is the thing that makes hu
20.
by FeepingCreature 22d ago
Though I also recommend reading the talk page, "weighted towards the truth" is not how I'd describe it. Personally I found the debate on the Jordan Lasker talk page very enlightening. Mother Jones is of course a very reliable
21.
by FeepingCreature 22d 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.
22.
by FeepingCreature 23d 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.
23.
by FeepingCreature 23d 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.
24.
by FeepingCreature 23d 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
25.
by FeepingCreature 24d ago
Strongly doubt that a LLM would say "domains that has .web". Maybe it's a human paraphrasing a LLM.
26.
by FeepingCreature 24d ago
I don't think that is true. It would be extremely noticeable (zero caching) and I am not noticing it.
27.
by FeepingCreature 24d ago
No, opencode allows bash by default.
28.
by FeepingCreature 1mo 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.
29.
by FeepingCreature 1mo 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.
30.
by FeepingCreature 1mo ago
Well great! You are already protected from random extensions then.

More ›