killerstorm

5 ms·

1.
A "slowdown" might improve OpenAI and Anthropic products(killerstorm.github.io)
2 points by killerstorm 3d ago|4 comments
2.
by killerstorm 3d ago
27% more efficient than base32 at expense of all kind of weird compat issues with shell scripts, etc. Yikes!
3.
by killerstorm 3d ago
You're confusing ToS guardrails with instruction-following issues and cheating. If a model fucks up your tests to report a success, it's not alligned.
4.
by killerstorm 5d ago
boys will be boys
5.
by killerstorm 6d ago
"Generalization" is rather unspecific. Analogy-making points to more specific mechanism: ability to identify features within representations, and modulating associative lookup using those features. I.e. it's not as simple as
6.
by killerstorm 6d ago
When my daughter was 6 months old, she once pointed at a towel and said "P". Apparently she recognize Winnie the Pooh from a book she saw earlier. If you think about it, this requires recognizing something common in things which a
7.
by killerstorm 6d ago
MacOS has no problem with hibernation because hardware and software are designed together: it uses TPM to manage swap encryption. I assume "BIOS" is also optimized to enable fast boot on sleep. Linux can work with TPM but it'
8.
by killerstorm 8d ago
Hmm, can't we just keep a better track of who comes up with some idea/concept/problem, etc, not just "who published a proof"? It seems like new age requires a new form of publication instead of sticking to 17th cent
9.
by killerstorm 10d ago
On a related note ''My robot bought illegal drugs' (2015) https://www.bbc.com/future/article/20150721-my-robot-bought-...
10.
by killerstorm 11d ago
I don't think there's any valid reason to _require_ JS on a static web site unless interactivity is the whole point.
11.
by killerstorm 12d ago
I tried it with some humanities questions and with the default OR system prompt (no prompt?) it seems to be rather mild - lacking usual AI mannerisms. Kinda cool.
12.
by killerstorm 12d ago
Xanadu principle have outlined important problems. This does not mean that Xanadu created practical solution, though: it seems they were focused on text transclusions UI for some reason... If the goal of Xanadu was to offer a better, more p
13.
by killerstorm 13d ago
Something like that. I think DNS registry system works pretty well under ICANN supervision: user isn't tied to a specific registrar.
14.
by killerstorm 14d ago
The government haven't noticed that email became as important as regular mail, unfortunately.
15.
by killerstorm 14d ago
That's just over-fit. Happens in humans too - like an alcoholic might swear to stop drinking. It was demonstrated that LLMs are capable of non-trivial self-introspection. E.g. if a steering vector is injected into residual stream, a su
16.
by killerstorm 14d ago
Perhaps "true" self-referentiality was not needed. But it seems "dynamic feedback" is essential, and it seems to be adjacent to self-referentiality. If we look at optimization process, it first does a forward pass which
17.
by killerstorm 14d ago
A lot of language nuance can only be picked up through cultural immersion, e.g. seeing how native speakers interact. (Which is a lot easier nowadays via Netflix, Youtube, etc.) It's pretty much impossible to "speak your language w
18.
by killerstorm 15d ago
Eh? STE is designed for procedural writing, e.g. "if you have a problem, press this button". I generally don't ask advanced AI to give me step-by-step instructions.
19.
by killerstorm 15d ago
This video actually explains math reasoning on a high level. Have you actually watched it?
20.
by killerstorm 17d ago
Rationalists generally prescribe a 'Bayesian' world-view, which can extract useful information out of a chaotic, too-complex-to-model world. Regarding the power of intelligence, it's generally considered to be synonymous with
21.
by killerstorm 17d ago
What "sci-fi baggage"? "Intelligence explosion" was first described by I.J. Good, a mathematician. von Neumann described singularity as a result of accelerating technical progress. He's also a mathematician, not a s
22.
by killerstorm 17d ago
When the "LessWrong crowd" talks about dangers of AI, they don't assume a particular form of intelligence or method to achieve it. They talk about the danger of optimization processes, i.e. "find X which minimize Y(X)&qu
23.
by killerstorm 17d ago
Oh, where are the AI unsafety organizations to provide us with truly unbiased investigations...
24.
by killerstorm 23d ago
Fun fact: in Soviet Union a common drug which people believe might reduce chance of a heart attack or stroke in high-stress/panic situation was Validol. What people don't know is that it was just peppermint oil + glucose. I.e. a m
25.
by killerstorm 27d ago
This remind me kind of stuff people did back in 2023: https://x.com/SigGravitas/status/1642181498278408193 ... Massive Update for Auto-GPT: Code Execution! Auto-GPT is now able to write it's own code using #g
26.
by killerstorm 27d ago
"The bigger they are, the harder they fall." If you're smart you probably have bigger, more challenging goals, so chances of success aren't higher. Deliberately setting goals lower might not help. Having a mindset which
27.
by killerstorm 28d ago
MIDI is still widely used for professional music production. It sounded goofy back in the day because synthesizer it was played on was not very good.
28.
by killerstorm 29d ago
Suppose you came up with AI algorithm which is better than transformers and such. Your options are: 1. publish it e.g. as a paper or open source, letting all big AI labs to use it for free 2. keep it secret Are you saying that it's evi
29.
by killerstorm 29d ago
> the guy in first place has to spend all that time and effort to develop the thing The problem is that patenting something doesn't require one to put a lot of time and effort - just to produce few pages of text and diagrams. So it
30.
by killerstorm 1mo ago
The question is whether it can accelerate scientific research, not whether it can completely replace human creativity.

More ›