Hacker News: zone411

New comment by zone411 in "OpenAI's GPT-6 Astra on ARC-AGI-3"

zone411 — Fri, 04 Sep 2026 03:23:14 +0000

I don't:

"Problems solved before a model's training cutoff can be filtered out, and all models compared on the remaining problems" means that the problems an older model actually solved are the ones that get filtered out, while the remaining problems are the ones it already tried and failed on. So older models end up with 0s on the filtered set and you can't really use this to compare new models to older ones.

Also, since these are known public problems, you can't stop people from spending far more than your arbitrary time and $ limits on them. So the number of clean problems will go down over time.

New comment by zone411 in "Intelligence Is Not the Main Bottleneck"

zone411 — Wed, 05 Aug 2026 16:16:55 +0000

A complete mischaracterization, as usual for HN lately when discussing AI or LessWrong. Obviously, even average levels of persuasion are enough to convince some people. And nobody is air-gapping AI.

New comment by zone411 in "Be skeptical of OpenAI's rogue hacker agent story"

zone411 — Sat, 25 Jul 2026 04:46:44 +0000

So why don't companies in other industries rush to prove their products are dangerous weapons? Maybe because it would be a really dumb PR stunt?

New comment by zone411 in "Be skeptical of OpenAI's rogue hacker agent story"

zone411 — Sat, 25 Jul 2026 04:44:59 +0000

And how do people saying this know the capabilities of yet unreleased models?

New comment by zone411 in "Be skeptical of OpenAI's rogue hacker agent story"

zone411 — Sat, 25 Jul 2026 04:43:29 +0000

How is it in their interest? Scaring customers, worrying employees, and inviting regulators to act is in their interest?

New comment by zone411 in "Be skeptical of OpenAI's rogue hacker agent story"

zone411 — Sat, 25 Jul 2026 04:41:29 +0000

It's not good marketing for them. This is a talking point with zero evidence that people repeat mindlessly. Scaring customers, worrying employees, and inviting regulators to act would be the worst marketing idea ever devised.

Natural-Density Almost-Bounded Collatz Orbits in Logarithmic Time (AI, Lean)

zone411 — Tue, 21 Jul 2026 19:18:43 +0000

Article URL: https://www.proofatlas.ai/formalizations/natural-density-log-time-collatz/

Comments URL: https://news.ycombinator.com/item?id=48996871

Points: 2

# Comments: 0

New comment by zone411 in "Openrouter Fusion API"

zone411 — Mon, 15 Jun 2026 17:16:25 +0000

Yes, definitely not a new idea. I had a multi-turn composite model in 2024 that was outperforming the top models across benchmarks: https://x.com/LechMazur/status/1828804485033992514.

New comment by zone411 in "Artificial intelligence is not conscious – Ted Chiang"

zone411 — Thu, 04 Jun 2026 17:16:07 +0000

That's not proof. Emergent intelligence is not consciousness.

New comment by zone411 in "The mysterious Hy3 LLM is topping OpenRouter Model Rankings by a large margin"

zone411 — Fri, 29 May 2026 05:10:19 +0000

I’ve tested this model on four of my benchmarks:

https://github.com/lechmazur/buyout_game 10th out 36.

https://github.com/lechmazur/pact/ 14th out 25.

https://github.com/lechmazur/nyt-connections/ 60th out 81.

https://github.com/lechmazur/debate 16th out of 29.

New comment by zone411 in "Spain blocks prediction markets Polymarket, Kalshi over lack of gambling licence"

zone411 — Tue, 26 May 2026 18:22:10 +0000

100%. It's sad to see that this attitude has spread to HN

New comment by zone411 in "An OpenAI model has disproved a central conjecture in discrete geometry"

zone411 — Wed, 20 May 2026 23:18:59 +0000

I actually tried using GPT-5.5 Pro on this problem recently. It thought it was making progress on one path, but it made so many mistakes that it didn't feel worth it pushing further. It'll be interesting to check whether it's the same route. I got partial results (proved in Lean) that improve on the best-known results for four Erdős problems with GPT-5.5 Pro

LLM Position Bias Benchmark: Swapped-Order Pairwise Judging

zone411 — Tue, 21 Apr 2026 22:08:12 +0000

Article URL: https://github.com/lechmazur/position_bias

Comments URL: https://news.ycombinator.com/item?id=47855226

Points: 1

# Comments: 0

New comment by zone411 in "EFF is leaving X"

zone411 — Thu, 09 Apr 2026 18:25:49 +0000

https://variety.com/2020/digital/news/twitter-unblocks-new-y...

Show HN: Buyout Game Benchmark: Multi-Agent Bargaining, Transfers, and Takeovers

zone411 — Mon, 30 Mar 2026 22:07:13 +0000

Article URL: https://github.com/lechmazur/buyout_game

Comments URL: https://news.ycombinator.com/item?id=47580319

Points: 6

# Comments: 0

New comment by zone411 in "AI overly affirms users asking for personal advice"

zone411 — Sat, 28 Mar 2026 16:51:13 +0000

I built this benchmark this month: https://github.com/lechmazur/sycophancy. There are large differences between LLMs. There are large differences between LLMs. For example, Mistral Large 3 and GPT-4.1 will initially agree with the narrator, while Gemini will disagree. I swap sides, so this is not about possible viewpoint bias in the LLMs. But another benchmark shows that Gemini will then change its view very easily in a multi-turn conversation while Kimi K2.5 or Grok won't: https://github.com/lechmazur/persuasion.

New comment by zone411 in "Folk are getting dangerously attached to AI that always tells them they're right"

zone411 — Sat, 28 Mar 2026 16:45:52 +0000

I built two related benchmarks this month: https://github.com/lechmazur/sycophancy and https://github.com/lechmazur/persuasion. There are large differences between LLMs. For example, good luck getting Grok to change its view, while Gemini 3.1 Pro will usually disagree with the narrator at first but then change its position very easily when pushed.

LLM Persuasion Benchmark: Multi-Turn Persuasion Between Models

zone411 — Fri, 27 Mar 2026 17:02:35 +0000

Article URL: https://github.com/lechmazur/persuasion

Comments URL: https://news.ycombinator.com/item?id=47545308

Points: 9

# Comments: 0

New comment by zone411 in "Show HN: LLM Debate Benchmark"

zone411 — Mon, 23 Mar 2026 21:10:35 +0000

Hmm, maybe in the next edition, Opus gets expensive. I should probably run GPT-5.4 xhigh too if I do that for fairness...

Show HN: LLM Debate Benchmark

zone411 — Mon, 23 Mar 2026 20:49:45 +0000

Article URL: https://github.com/lechmazur/debate/

Comments URL: https://news.ycombinator.com/item?id=47494895

Points: 9

# Comments: 3