152334H

7 ms·

31.
by 152334H 8mo ago
holy crap, this is so good. How did it get buried?
32.
Can AI agents write kernel exploits?(152334h.github.io)
3 points by 152334H 8mo ago|0 comments
33.
by 152334H 8mo ago
Thank you for the discovery.
34.
by 152334H 9mo ago
in effect, broadly anti-AI communities like bsky succeed by the sheer power of universal hate. Social policing can get you very far without any technology I think
35.
by 152334H 9mo ago
> I’ve never met a person saying they hate books and wish they were white on black. Wow. This is the most insightful statement I've read today. It's finally clicked [U+2014] why I prefer e-ink to hardback, in spite of the allur
36.
by 152334H 9mo ago
It's branched from how I think about TFR. To merge, - the common sentiment of "child raising is too expensive" - the reality that wealth has drastically gone up I think: okay, it must feel expensive for some reason. Probabl
37.
by 152334H 9mo ago
From my perspective, the issue is quite simple: progress optimizes everything other than the cost of human labor. Socialization, as defined today, inherently requires human labor, and thus falls under the Baumol effect. The 'loneliness
38.
by 152334H 9mo ago
It is a bit misleading to say 'Solo', when TinyTapeout is involved. Still clearly effortful work, though. I don't want to disparage it.
39.
by 152334H 9mo ago
good observations of AIGC form here Continually sad to see each day that the vast majority of users cannot classify
40.
by 152334H 9mo ago
oai/gdm avoids the issue on their public facing products with broader default moderation thresholds, which can still be turned down explicitly via their APIs. xai's only failure was to implement this modicum of damage control agai
41.
by 152334H 10mo ago
note that the first chunk of the piece spends time to analogize SV to the CCP, in terms of its willingness to take attacks (of humor). So, for your quote, a skeptical interpretation of the text may assert the author was merely praising SV i
42.
by 152334H 10mo ago
> Beijing has been preparing for Cold War without eagerness for waging it, while the US wants to wage a Cold War without preparing for it. great line
43.
by 152334H 10mo ago
thank you. absurd no other comments noticed
44.
by 152334H 10mo ago
+1. The article itself hides it well, but the draft paper linked at the end is clearly 100% AIGC. And the paper is clearly the ancestor to the article itself, based on the date (5dec -> 11dec)
45.
by 152334H 1y ago
amazing meta-level joke spanning months of effort... all articles on the author's site are themselves AI generated, and the article purporting to bemoan the impact of AI is itself AI Generated Content
46.
by 152334H 1y ago
what's supposed to be inferred from the link? I only see the homepage.
47.
by 152334H 1y ago
Extremely funny post. The author doesn't grasp how much of what they've written amounts to flexing their own outlier intelligence; they must sincerely believe the average programmer is capable of juggling a complex 500 line progra
48.
by 152334H 1y ago
They have a 20b for GPU poors, too. I will be running the 120B on my 2x4090-48GB, though.
49.
by 152334H 1y ago
'was'. I departed almost half a year prior to v3's release this week.
50.
by 152334H 2y ago
VP of AI at FAIR , who is unrelated to the llama/genai team.
51.
Llama 4 Computer Use Agent(llama4.pages.dev)
1 points by 152334H 2y ago|0 comments
52.
by 152334H 2y ago
Is it free-priority based? I was told by an employee that GDM internally has a credits system for TPU allocation, with which researchers have to budget out their compute usage. I may have completely misunderstood what they were describing,
53.
by 152334H 2y ago
You are correct on true H100 ownership costs being far lower. As I mention in the H100 blurb, the H100 numbers are fungible and I don't mind if you halve them. MFU can certainly be improved beyond 40%, as I mention. But on the point of
54.
Calculating the cost of a Google DeepMind paper(152334h.github.io)
303 points by 152334H 2y ago|150 comments
55.
by 152334H 2y ago
thanks
56.
Knowing Enough About MoE to Explain Dropped Tokens in GPT-4(152334h.github.io)
3 points by 152334H 3y ago|1 comments
57.
by 152334H 3y ago
https://www.youtube.com/watch?v=iFImKMjM-q4
58.
by 152334H 3y ago
just set up a desktop service to randomly open a paper once every few hours if they're not too boring, and you're not doing anything important, you'll read it for fun
59.
by 152334H 3y ago
I agree, and really hope that Meta is doing something in that vein. Reducing the FLOPs:Memory ratio (as in Soft MoE) could also open the door to CPU (or at least Apple Silicon) inference becoming more relevant.
60.
by 152334H 3y ago
Thanks. I'm really no expert (:P) on MoE research; I just noticed what was written in the Soft MoE paper and felt a need to check. The non-deterministic outputs are really similar, yeah, if you check the gist examples I linked https:&

More ›