Chu4eeno

8 ms·

91.
by Chu4eeno 3mo ago
Yeah, Anthropic likely is gaining an edge in tests like this from the data they got from Canva.
92.
by Chu4eeno 3mo ago
There's a ton of arenamaxxing going on (especially from facebook), though I don't disagree that it's one of the better actual benchmarks. Always fun to ask them to recreate classic demoscene effects (sadly they're still
93.
by Chu4eeno 3mo ago
The problem is that everyone (well, too many) is now using the same "voice" when writing. So no matter how good or thoughtful the writing is it gets tiresome.
94.
by Chu4eeno 4mo ago
There seems to be a ton of agents (openclaw or similar) set up instructed to just loop creating github projects or blogs and submit them to HN (optionally with a comment). Several of them make/write about the same thing as well, so I w
95.
by Chu4eeno 4mo ago
No, but it's extremely common. Whenever I've met people who claim to have "reasoned" their way out of religion it has always been extremely shallow teenage rebellion, and always driven by feelings. Even aalewis was &quot
96.
by Chu4eeno 4mo ago
That's usually the biggest problem with slop, the LLM is just expanding a short prompt into a full article. So it could just have been a single prompt, not even half the length.
97.
by Chu4eeno 4mo ago
No it didn't, the ukrainian flag was a recent addition by the guy who took over xsnow to add this protestware. Just go look at the repo history on sf.
98.
by Chu4eeno 4mo ago
Everyone that oppose him that I look into have conflicting interests, which makes djb's point even more concerning. The person you linked to works for Google, which he seems to omit mentioning in the mail you linked to.
99.
by Chu4eeno 4mo ago
I'm pretty sure the majority of content on reddit now is from bots as well (I used to moderate a medium sized subreddit pre-LLM, and the amount of automated posting was concerning even then). There's a ton of bots here as well (es
100.
by Chu4eeno 4mo ago
No, it literally is dumping. E. g. Chinese EV makers were selling at a loss (even with subsidies) until very recently this year and are struggling immensely now.
101.
by Chu4eeno 4mo ago
Goddamn timeless, well worth the read.
102.
by Chu4eeno 4mo ago
from the screenshots it looks like manual data entry
103.
by Chu4eeno 4mo ago
> That word 'some' does quite a bit of heavy lifting. No, I was just indicating I was not a sith.
104.
by Chu4eeno 4mo ago
https://www.theinformation.com/articles/xai-bets-groks-racy-... seems to be the original (paywalled) article, anyone have access or know how they gathered the data?
105.
by Chu4eeno 4mo ago
tldr (since they write extremely verbosely and meandering, and the link was tiny gray on gray): https://github.com/libunwind/libunwind/commit/a9b9293b286c14... https://github.com/libunwind&#x2
106.
by Chu4eeno 4mo ago
Did you confirm the verification with your own eyes? The verification report `bgagent` committed is very typical for hallucination (emojis, detailed timings, bold everywhere).
107.
by Chu4eeno 4mo ago
From the network it seems like a ton of people have just forked for themselves, plus a ton of unanswered issues and PRs, so I assume it's abandoned: https://github.com/gregmalcolm/python_koans/network
108.
by Chu4eeno 4mo ago
I wonder how much of this is vibe coded? Didn't see any estimate of that from skimming. It's a bit surprising, Claude, ChatGPT, Gemini and even Grok writes extremely robust and defensive C and C++ when I test them (various custom
109.
by Chu4eeno 4mo ago
They aren't necessarily worse and more unpredictable. IIRC. hallucination levels for the frontier levels are below equivalent average normal human errors, and they might even be more systematic/predictable to some degree. But even
110.
by Chu4eeno 4mo ago
Pretty impressive, though also a bit messy (and large), mixing glib, gtk, qt and some pure C. Plus some dead code (src/bytecode_cache.{c,h} at least). But honestly much cleaner and seems much more complete at first glance than the &quo
111.
by Chu4eeno 4mo ago
Is this the same company that just retracted a model after people got it to say some rather unsavory things about e. g. Anne Frank?
112.
by Chu4eeno 4mo ago
I believe grok embeds some modern standard provenance metadata (haven't checked what data it contains though), and Google uses their own invisible watermarking (that unfortunately only google can detect...).
113.
by Chu4eeno 4mo ago
It is technically impossible. People are already selling Worldcoin and other ID verifications for bot accounts for pretty cheap online (seems to be mostly targeted towards chinese users that want access to Claude).
114.
by Chu4eeno 4mo ago
I'm not him, but it creates some perverse incentives (like chinese billionaires who pay american surrogates to get implanted with dozens of their kids who get american citizenship). https://fortune.com/article/chin
115.
by Chu4eeno 4mo ago
It has extremely low false positive rate in various tests (e. g. https://arxiv.org/html/2501.15654v1 ), I've never seen it mark something I knew was human written as AI.
116.
by Chu4eeno 4mo ago
On Linux you can delete all hardlinks to open files. What you can't do is modify a file that's executing (program or library), then you get ETXTBSY.
117.
by Chu4eeno 4mo ago
It's a bit of a lost cause, go to /newest sometime, there's like 20 of these every day (LLM generated repo, LLM generated post/title, LLM generated description in comment, almost always LLM related porject, luckily often
118.
by Chu4eeno 4mo ago
They have always sent a ton of analytics + everything you do, is this news just because the Chinese are afraid they'll get cut off?
119.
by Chu4eeno 4mo ago
So they used to be just reselling access to «"Nemo, OpenHands 32B, OLMO 2 32B, and Mistral Small 3". OpenHands is a QWEN fine-tune, and Nemo and Mistral Small are both Mistral models.» But seems like they stripped this info from t
120.
by Chu4eeno 4mo ago
Because search engines suck, it's hard to find reliable information on relatively niche foreign news sources and I'm lazy.

More ›