mustaphah

5 ms·

31.
by mustaphah 7mo ago
That's a fair clarification. I never said we have sufficient evidence to act. But "too good to be true" + "singular paper" together can become an unfalsifiable dismissal - by that logic, every important result looks
32.
by mustaphah 7mo ago
I guess "too good to be true" is not a decent argument to convince a rational mind
33.
Tell HN: Claude two rate limits don't know about each other
2 points by mustaphah 7mo ago|0 comments
34.
by mustaphah 7mo ago
> After you read a lot of those you learn not to take individual papers seriously Can't disagree, but keep in mind that almost all meds are tested first in mice/animal models before human trials verify the effects.
35.
by mustaphah 7mo ago
I guess it's because most major disorders and diseases have so many pathways at play that figuring out which one's actually causing the problem at the individual level is just too tricky. The other thing concerns how potent the ef
36.
by mustaphah 7mo ago
Chili could work [1], but not too much of it [2] [1] https://pubmed.ncbi.nlm.nih.gov/27079706/ [2] https://pubmed.ncbi.nlm.nih.gov/31137805/
37.
by mustaphah 7mo ago
Yeah, it's a mouse study, but there are tons of human studies backing the whole gut-brain connection. There are even a bunch of books on it [1][2]. What's really cool is that the paper used low-dose capsaicin (just 5 μg/kg in
38.
Enhancing gut-brain communication reversed cognitive decline in aging mice(med.stanford.edu)
386 points by mustaphah 7mo ago|185 comments
39.
Many SWE-bench-Passing PRs would not be merged(metr.org)
278 points by mustaphah 7mo ago|153 comments
40.
by mustaphah 8mo ago
Well, it's fixed now. Apparently, Claude Code was still up and running!
41.
by mustaphah 8mo ago
I'd bet it's Cloudflare's fault
42.
AGI is an unscientific myth(tandfonline.com)
4 points by mustaphah 8mo ago|2 comments
43.
by mustaphah 8mo ago
Even worse, I got contacted through YC Jobs (workatastartup.com) with a message that was basically: "Star, fork, and submit PRs to our open-source repo and we'll review you for a contract." I immediately realize it's eng
44.
by mustaphah 8mo ago
Worth noting: emails from .online domain (and many other TLDs [1]) are also way more likely to end up in the spam folder. https://www.spamhaus.org/reputation-statistics/gtlds/malware...
45.
Web Verbs(github.com)
1 points by mustaphah 8mo ago|0 comments
46.
by mustaphah 8mo ago
That's been my experience too; can't disagree. Still, when it comes to tasks that require deep intelligence (esp. mathematical reasoning [1]), Gemini has consistently been the best. [1] https://arxiv.org/abs/2
47.
by mustaphah 8mo ago
Google is terrible at marketing, but this feels like a big step forward. As per the announcement, Gemini 3.1 Pro score 68.5% on Terminal-Bench 2.0, which makes it the top performer on the Terminus 2 harness [1]. That harness is a "neut
48.
OpenAI's 5-month experiment: building a product with no human-written code(openai.com)
2 points by mustaphah 8mo ago|0 comments
49.
by mustaphah 8mo ago
I'm fine with editing the title, and I see your point. In retrospect, I'd probably avoid "useless." While it's a fairly descriptive term for their finding, it probably carries a subjective tone.
50.
by mustaphah 8mo ago
Yeah, I got your point when I read the paper. You're essentially controlling for "latent domain knowledge." I might have been a bit blunt with the title - sorry about that, but I still think it was a good title. From what I&#
51.
by mustaphah 8mo ago
This is like trying to fix hallucination by telling LLM not to hallucinate.
52.
by mustaphah 8mo ago
What if the most interesting finding ends up buried under a vague title? Aside from the "self-generated skills" aspect, there isn't much there that meaningfully warrants deeper discussion. I chose a title that directly reflec
53.
SkillsBench: Benchmarking how well agent skills work across diverse tasks(arxiv.org)
364 points by mustaphah 8mo ago|171 comments
54.
Evaluating AGENTS.md: are they helpful for coding agents?(arxiv.org)
232 points by mustaphah 8mo ago|161 comments
55.
Curosr: Expanding our long-running agents research preview(cursor.com)
3 points by mustaphah 8mo ago|0 comments
56.
Measuring Time Horizon Using Claude Code and Codex(metr.org)
1 points by mustaphah 8mo ago|0 comments
57.
SWE-ContextBench: context learning benchmark in coding(arxiv.org)
1 points by mustaphah 8mo ago|0 comments
58.
by mustaphah 8mo ago
Leaderboard: https://swe-agi.com/
59.
SWE-AGI: benchmarking spec-driven software construction(arxiv.org)
1 points by mustaphah 8mo ago|1 comments
60.
Code Formatting Silently Consumes Your LLM Budget(arxiv.org)
1 points by mustaphah 8mo ago|0 comments

More ›