jmmcd

6 ms·

1.
by jmmcd 6d ago
Can't believe people are advocating free love in 2026
2.
by jmmcd 29d ago
AC at 68F is obnoxious and wasteful. At least data centres enable me to communicate with distant colleagues, watch enriching movies, and use very strong AI tools to accelerate my research.
3.
by jmmcd 3mo ago
A lot of us went through a similar system and still retained both honour and desire to actually learn.
4.
by jmmcd 3mo ago
It's interesting that Claude also over-uses en -dashes. It's very willing to create compound-noun-phrases, especially in that compressed-summary-paragraph it often writes. The 0-days-vibes-vulns that started this thread looks a
5.
by jmmcd 3mo ago
People are missing that Willison is among the very best people we have in the role of (for lack of a good name): early access to frontier models, evaluate them in real scenarios, no wishful thinking, hype, or doom, communicate the possibili
6.
by jmmcd 4mo ago
> LLMs do just interpolate their training data "interpolate" has a technical meaning - in this meaning, LLMs almost never interpolate. It also has a very vague everyday meaning - in this meaning, LLMs do interpolate, but so do
7.
by jmmcd 5mo ago
But euros spent on tokens is a tiny fraction of the overall costs of the project.
8.
by jmmcd 5mo ago
> Since these companies can’t improve their AI models without fresh data created by human beings Totally wrong. Self-play dates back to Arthur Samuel in the 1950s and RL with verifiable rewards is a key part of training the most advanced
9.
by jmmcd 5mo ago
> Anglo Please, write US-American. These people are not coming from any other place.
10.
by jmmcd 7mo ago
I understand your point, but in response to GP (they should spend this money on houses for other poor people instead), the reduced reliance on other social welfare is totally legitimate to count.
11.
by jmmcd 7mo ago
You didn't read the article. The scheme gave positive return on investment.
12.
by jmmcd 8mo ago
Google Scholar provides imperfect citations - very often wrong article type (eg article versus conference paper), but up to and including missing authors, in my experience.
13.
by jmmcd 8mo ago
Jokes are one of the good parts of human existence, so - while I see your side of the story - there is another side.
14.
by jmmcd 8mo ago
The best example of all is Prolog. It is always held up as the paradigmatic representative of logic programming, a rare language paradigm. But it doesn't need to be a language. It is really a collection of algorithms which should be a
15.
by jmmcd 10mo ago
But he actually uses frontier LLMs in his own work. Probably that's stronger evidence.
16.
by jmmcd 10mo ago
Yes. But "the gold standard" just means "the most natural, easy and dumb way".
17.
by jmmcd 10mo ago
This is on the semi-private set * https://x.com/arcprize/status/1990820655411909018 * https://arcprize.org/guide
18.
by jmmcd 10mo ago
"Pelican on bicycle" is one special case, but the problem (and the interesting point) is that with LLMs, they are always generalising. If a lab focussed specially on pelicans on bicycles, they would as a by-product improve perform
19.
by jmmcd 10mo ago
About ARC 2: I would want to hear more detail about prompts, frameworks, thinking time, etc., but they don't matter too much. The main caveat would be that this is probably on the public test set, so could be in pretraining, and there
20.
by jmmcd 1y ago
On HN it's very common to see a blog post along the lines of "I found this old piece of equipment with no brand name, I used some network traffic inspection to figure out what it does, I hacked around a bit, I got it working and t
21.
by jmmcd 1y ago
(a) no it's not (b) your comment is miles off-topic, as he is not addressing doom in any sense
22.
by jmmcd 1y ago
But they're submissions to ICML.
23.
by jmmcd 1y ago
It could certainly replace the author of this article.
24.
by jmmcd 1y ago
Modern LLMs can one-shot code in a totally new language, if you provide the language manual. And you have to provide the language manual, because otherwise how can the students learn the language.
25.
by jmmcd 1y ago
> The fact that AI can do your homework should tell you how much your homework is worth. A lot of people who say this kind of thing have, frankly, a very shallow view of what homework is. A lot of homework can be easily done by AI, or by
26.
by jmmcd 1y ago
Absolutely devastating for the credibility of FAIR.
27.
by jmmcd 1y ago
Link [1] doesn't seem to mention svg or vector graphics at all.
28.
by jmmcd 1y ago
This important paper from Anthropic includes evidence that part (but only part) of reasoning is cross-lingual: https://www.anthropic.com/research/tracing-thoughts-language...
29.
by jmmcd 1y ago
No, you're confusing GA with GP.
30.
by jmmcd 1y ago
> in genetic programming the goal is to find/optimize dataset to fit given algorithm No. Possibly you're confused between GAs and GP, a common confusion. In GP, the goal is to find an algorithm - in a real programming language,

More ›