killerstorm

3 ms·

1.
by killerstorm 5d ago
Making model to consists of many small modules is inefficient on GPU, especially as routing adds data dependencies, etc, and especially with pytorch (compared to a custom kernel). The difference might be smaller on a CPU which has limited p
2.
by killerstorm 6d ago
True, but I think in the end people were better off with automation. You know a lot of people were suffering even when they were fully employed.
3.
by killerstorm 6d ago
Luddites also complain about disruption, you know
4.
by killerstorm 6d ago
I'd say this social contract which got "broken" never existed in the first place. Even before AI we heard lots of complaints like "I made a popular open source library which is now used by corps with trillion-dollar mark
5.
by killerstorm 8d ago
> Look like how AI slop has unrealistic physics > Posts a link to real moon landing footage I'd delete the article if I was you... You know, in academia, they sometimes retract articles, even if they believe they are directionall
6.
by killerstorm 8d ago
This is, basically, text-to-LoRA with some extra stuff. I.e. it basically takes text, computes and embedding and makes a LoRA adapter out of this embedding. Note that it is equivalent to a recurrent module attached to a transformer. Dynamic
7.
by killerstorm 8d ago
Here's what Victor wrote about inets in Bend2 (on X): > interaction combinators still parallelize better than anything else, but the graph overhead prevents us from compiling to maximally efficient assembly. bend2 is basically inets
8.
by killerstorm 8d ago
A lot of information here: https://gist.github.com/VictorTaelin/77fd5a2a8a4a07e1da6157e... https://github.com/victortaelin
9.
by killerstorm 8d ago
A complete implementation have been released, how is that not a substantiation? Academic people might have more trust in a paper which when through a lengthy publication process. But if you think about it, it's not a better proof than
10.
by killerstorm 8d ago
That's a start-up style marketing: when you make a product you focus on a big vision and positive sides and de-emphasize weaknesses. I'm afraid that's actually 100% Victor's decision to do it this way, and it seems to b
11.
by killerstorm 8d ago
Victor put 5+ years of research into this. You can find many of previous versions (which use different approach, do a different kind of a thing, etc.) on the github. "Bend2" in particular have been in development for 2 years. Call
12.
by killerstorm 8d ago
Victor Taelin has been doing interesting PLT research for 10+ years. I suggest you read his history: https://gist.github.com/VictorTaelin/77fd5a2a8a4a07e1da6157e... before making slop accusations. Older variant of what
13.
by killerstorm 9d ago
Interesting that you mention LG and Samsung TVs. Does it happen on Google Pixel phones? Obviously, the quality of the walled garden depends on the maintainer. Google's quality standards are lower than Apples, but higher than LGs.
14.
by killerstorm 10d ago
Hmm? Malware on iOS is extremely rare. I remember in Bitcoin community ~10 years ago, standard recommendation was than an iOS wallet was secure enough (I don't recall even a single case where wallet was stolen via malware), but any pri
15.
by killerstorm 10d ago
Back in early 2000s, pretty much all Windows machines were infested with malware. Do you want to bring back those glorious days? Back in the day users didn't really have much valuable and sensitive stuff on their machines and malware w
16.
by killerstorm 10d ago
"Language Models are Few-Shot Learners" - 2020, the GPT-3 paper. It have been demonstrated that in-context learning is a very powerful mechanism. There's no evidence that models of the size of GPT-6 are bad at in-context lear
17.
by killerstorm 10d ago
This is an absolute nonsense. Any frontier model can implement chess program from scratch - modeling the board, checking legality, etc. If you asked e.g. GPT-6 to get good at chess and gave it a computer, it will get good at chess. That&#x2
18.
by killerstorm 12d ago
If AI is good at following instructions, desired behavior can be specified in a (system) prompt
19.
by killerstorm 12d ago
I doubt "slowdown" framing was optimal - could frame it as an acceleration towards more practical AI rather than deceleration
20.
A "slowdown" might improve OpenAI and Anthropic products(killerstorm.github.io)
2 points by killerstorm 12d ago|4 comments
21.
by killerstorm 13d ago
27% more efficient than base32 at expense of all kind of weird compat issues with shell scripts, etc. Yikes!
22.
by killerstorm 13d 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.
23.
by killerstorm 14d ago
boys will be boys
24.
by killerstorm 15d 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
25.
by killerstorm 15d 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
26.
by killerstorm 15d 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'
27.
by killerstorm 17d 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
28.
by killerstorm 20d ago
On a related note ''My robot bought illegal drugs' (2015) https://www.bbc.com/future/article/20150721-my-robot-bought-...
29.
by killerstorm 21d ago
I don't think there's any valid reason to _require_ JS on a static web site unless interactivity is the whole point.
30.
by killerstorm 21d 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.

More ›