sanxiyn

5 ms·

61.
by sanxiyn 7mo ago
There is. Official leaderboard is without harness, and community leaderboard is with harness. Read ARC-AGI-3 Technical Paper for details.
62.
by sanxiyn 7mo ago
That works well. Anthropic wrote a writeup on it. https://www.anthropic.com/engineering/harness-design-long-ru...
63.
by sanxiyn 7mo ago
Yes, but a programming language with a proverbial sufficiently smart compiler. That is very useful.
64.
by sanxiyn 7mo ago
While I think all of your design choices are defensible, I do think you should release the full human baseline data. The second best action count is fine, but other choices are reasonable as well.
65.
by sanxiyn 7mo ago
The study is likely "SWE-CI: Evaluating Agent Capabilities in Maintaining Codebases via Continuous Integration". Regression rate plot is figure 6. Read the study to understand what it is measuring and how it was measured. As I und
66.
by sanxiyn 8mo ago
You can disable this at Settings > Capabilities > Memory > Search and reference chats.
67.
by sanxiyn 9mo ago
Not really. Due to combinatorial explosion some path is hard to hit randomly in this kind of source code. I would have preferred if after 2M random battles the reference implementation had 99% code coverage, than 99% pass rate. I don't
68.
by sanxiyn 9mo ago
I think you are comparing latency with throughput. You can't take the inverse of latency to get throughput because concurrency is unknown. But then, RNGD result is probably with concurrency=1.
69.
by sanxiyn 9mo ago
Yes, but in principle it isn't that different from running on Trainium or Inferentia (it's a matter of degree), and plenty of non-AI organizations adopted Trainium/Inferentia.
70.
by sanxiyn 9mo ago
What's new is HN discovered it. It wasn't posted in September 2025.
71.
by sanxiyn 9mo ago
Looking at their blog, they in fact ran gpt-oss-120b: https://furiosa.ai/blog/serving-gpt-oss-120b-at-5-8-ms-tpot-... I think Llama 3 focus mostly reflects demand. It may be hard to believe, but many people aren't
72.
by sanxiyn 10mo ago
In my opinion, Groq's technical decisions are unsound in a normal world. But being HBM-free may have some merit in a world where HBM supply is constrained.
73.
by sanxiyn 10mo ago
Ruby also used to use Bison, uses its own https://github.com/ruby/lrama these days.
74.
by sanxiyn 11mo ago
Yeah. There are other fully open models like Hugging Face SmolLM but they are not common.
75.
by sanxiyn 11mo ago
With cargo --offline, Rust has better than average support for offline build.
76.
by sanxiyn 11mo ago
An interesting bit of history: for a long time Rust maintained first party support for Windows XP, after other parts of ecosystem generally gave up. This was because Firefox needed it. https://github.com/rust-lang/compi
77.
by sanxiyn 11mo ago
If you look at the actual code, it runs ping -c 5. I agree ping without options doesn't terminate.
78.
by sanxiyn 1y ago
Yes, but in fact compensating for bad questions is a skill, and in my experience it is a skill excelled by Claude and poorly by Gemini. In other words, better you are at prompting (eg you write a half page of prompt even for casual uses --
79.
by sanxiyn 1y ago
No, but with tail call (using underlying WebAssembly tail call).
80.
Generative design of novel bacteriophages with genome language models(biorxiv.org)
3 points by sanxiyn 1y ago|0 comments
81.
by sanxiyn 1y ago
> "We find that comments by GPT‑5-Codex are less likely to be incorrect or unimportant" -- less unimportant comments in code is definitely an improvement! This seems to be a misunderstanding. In the original OpenAI article, com
82.
by sanxiyn 1y ago
I am not sure how your sun example relates. Language is not whole of reality, but it is clearly part of reality. Memory engram of Coca-Cola is encoded in billions of human brains all over the world, and they are arrangement of atoms.
83.
by sanxiyn 1y ago
While the broad point that no, OpenAI and Anthropic are not losing money on inference, is almost certainly correct, the model itself is quite incomplete. If you actually want to know, I recommend Inference economics of language models from
84.
by sanxiyn 1y ago
Yes, models give likelihoods you can compare against. No, you can't do that without drafting, because likelihood of token N+2 depends on token N+1. That is, you get P(is, The capital of France) and P(Berlin, The capital of France is),
85.
by sanxiyn 1y ago
Terence Tao wrote on "blue team" vs "red team" in cybersecurity and how "unreliable" AI is more suited to red team side. I found it very insightful. https://news.ycombinator.com/item?id=44711306
86.
by sanxiyn 1y ago
I understand it is not your focus, but fuzzing still falls short and there is a lot AI can help. For example, when there is checksum, fuzzers typically can't progress and it is "solved" by disabling checks when building for f
87.
by sanxiyn 1y ago
Maybe their ISPs don't do that. There are many ISPs on the Earth.
88.
by sanxiyn 1y ago
You can look at Debian's reasoning here: https://wiki.debian.org/ArmHardFloatPort . As I understand, the decision was mostly based on hardware.
89.
by sanxiyn 1y ago
Google quote from https://riscv.org/riscv-news/2024/10/risc-v-announces-ratifi... > "Google is delighted to see the ratification of the RVA23 Profile," said Lars Bergstrom, Director of Engineer
90.
by sanxiyn 1y ago
In my experience long context approach flatly doesn't work, so I don't think this is it. The post does mention "tools for keeping notes and preserving important information to be checked later".

More ›