Hacker News: hooloovoo_zoo

New comment by hooloovoo_zoo in "Embarrassingly simple self-distillation improves code generation"

hooloovoo_zoo — Sat, 04 Apr 2026 22:18:13 +0000

They are training the model to 1. Produce code (as opposed to answer a question, write a poem, etc.) 2. Produce long enough output to be a valid solution. So they are doing exactly what I said. Cheers.

New comment by hooloovoo_zoo in "Embarrassingly simple self-distillation improves code generation"

hooloovoo_zoo — Sat, 04 Apr 2026 21:53:54 +0000

One sentence summary: We fine-tuned a general-purpose model to produce valid benchmark code results and it got better at producing benchmark code results; we didn't bother to evaluate it on anything the model used to be good at.

New comment by hooloovoo_zoo in "Claude Code Cheat Sheet"

hooloovoo_zoo — Tue, 24 Mar 2026 00:47:55 +0000

Proposition: Every power user feature added lowers Anthropic’s market cap $1B and OpenAI’s $10B.

New comment by hooloovoo_zoo in "Walmart fires OpenAI in playbook-changing move"

hooloovoo_zoo — Mon, 23 Mar 2026 00:50:27 +0000

If you ask the agent a product question you can trigger it into web search mode where it provides web links and autocomplete suggestions. You can follow the a/c suggestions down some pretty elaborate chains of questions but eventually it hits a rate limit and reverts to the default dumb bot.

New comment by hooloovoo_zoo in "Walmart fires OpenAI in playbook-changing move"

hooloovoo_zoo — Sun, 22 Mar 2026 20:02:15 +0000

Well, they seem to be using 3rd party LLMs for websearch.

New comment by hooloovoo_zoo in "Roblox is minting teen millionaires"

hooloovoo_zoo — Tue, 10 Mar 2026 21:58:29 +0000

Traffic for his game has declined by 90% from its peak already though.

New comment by hooloovoo_zoo in "Alleged Distillation Attacks by DeepSeek, Moonshot AI, and MiniMax"

hooloovoo_zoo — Mon, 23 Feb 2026 23:24:30 +0000

That would be an interesting paper actually; what is the optimal sampling technique given you only have access to the token outputs. Surely someone has already done it.

New comment by hooloovoo_zoo in "Alleged Distillation Attacks by DeepSeek, Moonshot AI, and MiniMax"

hooloovoo_zoo — Mon, 23 Feb 2026 23:19:05 +0000

Sampling repeatedly gives them an estimate of the probability distribution in any case though.

New comment by hooloovoo_zoo in "14-year-old Miles Wu folded origami pattern that holds 10k times its own weight"

hooloovoo_zoo — Mon, 16 Feb 2026 21:59:13 +0000

Looks kinda like an egg carton to me. So if an empty egg carton weighs 50g, that's like saying you could stack 500kg on top. Pretty impressive.

New comment by hooloovoo_zoo in "Show HN: I made a memory game to teach you to play piano by ear"

hooloovoo_zoo — Sat, 10 Jan 2026 01:24:13 +0000

Huh thanks that works but that’s not how silent mode is supposed to function. Still a bug imo.

New comment by hooloovoo_zoo in "Show HN: I made a memory game to teach you to play piano by ear"

hooloovoo_zoo — Fri, 09 Jan 2026 18:02:16 +0000

Don’t seem to get any sound on iPhone 16 even after clicking :(

New comment by hooloovoo_zoo in "DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning [pdf]"

hooloovoo_zoo — Fri, 28 Nov 2025 21:32:16 +0000

> Why do you think that the 2024 Putnam programs that they used to test were in the training data?

They reference https://artofproblemsolving.com/community/c13_contest_collec... for the source of their scrape and the Putnam problems are on that page under 'Undergraduate Contests'.

New comment by hooloovoo_zoo in "DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning [pdf]"

hooloovoo_zoo — Fri, 28 Nov 2025 07:41:16 +0000

For one thing, it's not a real score; they judged the results themselves and Putnam judges are notoriously tough. There was not a single 8 on the problem they claim partial credit for (or any partial credit above a 2) amongst the top 500 humans. https://kskedlaya.org/putnam-archive/putnam2024stats.html.

For another thing, the 2024 Putnam problems are in their RL data.

Also, it's very unclear how these competitions consisting of problems designed to have clear-cut answers and be solved by (well-prepared) humans in an hour will translate to anything else.

S.F. speed cameras are generating mountains of tickets and fines

hooloovoo_zoo — Fri, 03 Oct 2025 23:41:27 +0000

Article URL: https://www.sfchronicle.com/sf/article/speed-camera-tickets-data-21076408.php

Comments URL: https://news.ycombinator.com/item?id=45469053

Points: 13

# Comments: 7

New comment by hooloovoo_zoo in "OpenAI and Nvidia announce partnership to deploy 10GW of Nvidia systems"

hooloovoo_zoo — Mon, 22 Sep 2025 16:45:45 +0000

These $ figures based on compute credits or the investor's own hardware seem pretty sketchy.

New comment by hooloovoo_zoo in "Poisoning Well"

hooloovoo_zoo — Fri, 05 Sep 2025 18:03:39 +0000

My read is that they are describing functionality for site owners to provide input about what the site owner thinks should happen. OpenAI is not promising that is what WILL happen, even in the narrow context of that specific bot.

New comment by hooloovoo_zoo in "Poisoning Well"

hooloovoo_zoo — Fri, 05 Sep 2025 09:01:47 +0000

Your link does not say they will obey it.

New comment by hooloovoo_zoo in "No clicks, no content: The unsustainable future of AI search"

hooloovoo_zoo — Sun, 31 Aug 2025 21:09:52 +0000

It's just a new era in the ad-block wars where my ad-block LLM tries to excise the ads from their ad-generating LLM.

New comment by hooloovoo_zoo in "Meta Freezes AI Hiring After Blockbuster Spending Spree"

hooloovoo_zoo — Thu, 21 Aug 2025 02:06:03 +0000

Meta already bought all openAI’s secrets; now it just needs to let the GPUs cook.

New comment by hooloovoo_zoo in "LLMs tell bad jokes because they avoid surprises"

hooloovoo_zoo — Sun, 17 Aug 2025 18:47:31 +0000

Maybe popsicle stick funny 'It was burnt out with being a rock-star'.