Hacker News: piotrgrabowski

Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA

piotrgrabowski — Tue, 21 Jul 2026 22:35:24 +0000

Also: Kimi K3: second only to Fable 5 on AA-Briefcase https://artificialanalysis.ai/articles/kimi-k3-agentic-knowl...

Comments URL: https://news.ycombinator.com/item?id=48999291

Points: 877

# Comments: 449

Next 100 Days: XBOX Reset

piotrgrabowski — Wed, 10 Jun 2026 22:13:28 +0000

Article URL: https://news.xbox.com/en-us/2026/06/10/next-100-days-xbox-reset/

Comments URL: https://news.ycombinator.com/item?id=48483456

Points: 6

# Comments: 0

Opus 4.5 Collapsed Six Months of Development Work into One Week

piotrgrabowski — Thu, 04 Dec 2025 19:05:13 +0000

Article URL: https://every.to/chain-of-thought/opus-4-5-collapsed-six-months-of-development-work-into-one-week

Comments URL: https://news.ycombinator.com/item?id=46151457

Points: 2

# Comments: 0

Ilya Sutskever: We're moving from the age of scaling to the age of research

piotrgrabowski — Tue, 25 Nov 2025 17:21:52 +0000

Article URL: https://www.dwarkesh.com/p/ilya-sutskever-2

Comments URL: https://news.ycombinator.com/item?id=46048125

Points: 450

# Comments: 355

New comment by piotrgrabowski in "CompileBench: Can AI Compile 22-year-old Code?"

piotrgrabowski — Mon, 22 Sep 2025 13:57:20 +0000

Author here.

So far in this benchmark we based the tasks on a couple of open-source projects (like curl, jq, GNU Coreutils).

Even on those "simple" projects we managed to make the tasks difficult - Claude Opus 4.1 was the only one to correctly cross-compile curl for arm64 (+ make it statically-linked) [1].

In the future we'd like to test it with projects like FFmpeg or chromium - those should be much more difficult.

[1] https://www.compilebench.com/curl-ssl-arm64-static/

Show HN: 60-Second Linux Analysis, Supercharged with Nix and LLMs

piotrgrabowski — Sat, 06 Sep 2025 15:53:30 +0000

Hello HN,

I'm sharing a little open-source utility I wrote recently.

I'm a huge fan of Brendan Gregg's "BPF Performance Tools" book. However, every time I SSH into a fresh server, most of the diagnostic tools aren't installed there and installing them can be really annoying.

I decided to use Nix package manager and LLMs to make this process straightforward. My utility first downloads a "toolbox" of Linux utilities (built with Nix), runs Brendan Gregg's famous "60-second Linux analysis" playbook and then summarizes the results with an LLM. So "60-second Linux analysis" now becomes a single one-line command and actually takes less than 60 seconds! The utility can execute all commands in parallel and the LLM can analyze them faster than a human would.

I have a few ideas for the future, for example implementing more powerful playbooks - thanks to Nix I can easily bundle all tools I need and LLMs have no trouble analyzing outputs of tens of commands.

I'd love to get your feedback and hear any ideas you have. Thanks for checking it out.

You can launch the utility with this command:

$ curl -fsSL https://gradient.engineer/60-second-linux.sh | sh

Comments URL: https://news.ycombinator.com/item?id=45150357

Points: 4

# Comments: 1

Automatically Jailbreaking Frontier Language Models with Investigator Agents

piotrgrabowski — Wed, 03 Sep 2025 17:18:01 +0000

Article URL: https://transluce.org/jailbreaking-frontier-models

Comments URL: https://news.ycombinator.com/item?id=45118285

Points: 2

# Comments: 0