Hacker News: s314

New comment by s314 in "Ovid: A pi extension that makes it record proof its features actually work"

s314 — Wed, 01 Jul 2026 00:10:50 +0000

I built https://github.com/Srinivasa314/ovid, a pi extension that makes it verify the features it builds and record terminal+browser videos onto the PR. The verifications are ordinary code, so re-running them is cheap and no LLM is needed.

Ovid: A pi extension that makes it record proof its features actually work

s314 — Wed, 01 Jul 2026 00:10:50 +0000

Article URL: https://github.com/Srinivasa314/ovid

Comments URL: https://news.ycombinator.com/item?id=48740884

Points: 2

# Comments: 1

Porting Zircon (Fuchsia OS's kernel) C++ code to Rust

s314 — Sun, 07 Jun 2026 03:59:18 +0000

Article URL: https://cs.opensource.google/fuchsia/fuchsia/+/main:zircon/skills/cpp-to-rust-coder/SKILL.md

Comments URL: https://news.ycombinator.com/item?id=48431640

Points: 4

# Comments: 2

New comment by s314 in "What political censorship looks like inside an LLM's weights (Qwen 3.5)"

s314 — Tue, 19 May 2026 03:27:58 +0000

Using a logit lens (prior art: https://www.lesswrong.com/posts/AcKRB8wDpdaN6v6ru/interpreti...)

New comment by s314 in "What political censorship looks like inside an LLM's weights (Qwen 3.5)"

s314 — Tue, 19 May 2026 02:44:58 +0000

You can actually de-censor an LLM without understanding how it works from a mechanistic perspective. (See R1 1776)

So I don't think there'll be effort to "obfuscate"

New comment by s314 in "What political censorship looks like inside an LLM's weights (Qwen 3.5)"

s314 — Tue, 19 May 2026 02:26:52 +0000

It wasn't "a prompt" but several prompts that transformed the raw experimental results to a blog.

> hallucinated from the LLMs world knowledge

This can't be true because I checked whether the content was consistent with the experimental outputs

New comment by s314 in "What political censorship looks like inside an LLM's weights (Qwen 3.5)"

s314 — Tue, 19 May 2026 01:39:44 +0000

> Steering seems like a circumventable kludge compared to adjusting the training data directly

Correct. Steering is used in mechanistic interpretability studies to prove that your model is correct. There are other better ways to "decensor".

What political censorship looks like inside an LLM's weights (Qwen 3.5)

s314 — Tue, 19 May 2026 00:16:31 +0000

Article URL: https://vas-blog.pages.dev/qwen-censorship/

Comments URL: https://news.ycombinator.com/item?id=48187680

Points: 82

# Comments: 44

New comment by s314 in "Using Claude Code: The Unreasonable Effectiveness of HTML"

s314 — Sat, 09 May 2026 02:55:21 +0000

Markdown feels too minimal and HTML feels too complex. I think we need something in between

New comment by s314 in "A text adventure game, all in Markdown"

s314 — Wed, 22 Apr 2026 01:23:39 +0000

Expanding this analogy: LLMs like Claude are like V8, and agent harnesses like Claude Code are like Node.js.

New comment by s314 in "A text adventure game, all in Markdown"

s314 — Wed, 22 Apr 2026 01:18:01 +0000

You can think of the LLM like an interpreter.

A text adventure game, all in Markdown

s314 — Wed, 22 Apr 2026 00:23:12 +0000

Article URL: https://vas-blog.pages.dev/markdown-adventure

Comments URL: https://news.ycombinator.com/item?id=47856887

Points: 1

# Comments: 3

Crates.io: Malicious crates faster_log and async_println

s314 — Fri, 26 Sep 2025 08:47:14 +0000

Article URL: https://blog.rust-lang.org/2025/09/24/crates.io-malicious-crates-fasterlog-and-asyncprintln/

Comments URL: https://news.ycombinator.com/item?id=45384276

Points: 2

# Comments: 0