Hacker News: aesthesia

New comment by aesthesia in "Qwen 3.8 27B is excellent, but it defaults to overthinking things"

aesthesia — Mon, 17 Aug 2026 16:01:50 +0000

Has this hypothesis that reasoning tokens help the model engage a wider range of experts been tested? I'm a little skeptical that it's the main driver of extended reasoning traces, particularly because MoE models are generally already trained so that expert activations are as uniformly distributed as possible. But it would be interesting to know to what extent this happens.

New comment by aesthesia in "Timeline of the OpenAI accidental attack against Hugging Face"

aesthesia — Sat, 08 Aug 2026 20:51:57 +0000

What makes you think that this is actually good PR for the firms involved? Every claim that this is good PR comes from someone who has increased their negative views of OpenAI based on these events. Where are the people coming away with a positive impression? This seems like making up a guy to get mad at.

New comment by aesthesia in "Why Erdős Problems Are Falling to AI"

aesthesia — Wed, 05 Aug 2026 17:15:10 +0000

I'm not sure we should take it as a given that AI will not also become better than humans at mathematical exposition.

New comment by aesthesia in "Agent Skill to Force Docs in ASD-STE100 Simplified Technical English"

aesthesia — Fri, 31 Jul 2026 05:24:14 +0000

Ah, I see, taking "to test A" as an infinitive. But it would be strange to say test B is an alternative without saying what it's an alternative to. The other reading still seems quite unlikely.

New comment by aesthesia in "A California aquifer may have crossed the point of no return"

aesthesia — Fri, 31 Jul 2026 05:21:12 +0000

You can make this kind of claim about living in lots of different places. Some amount of engineering and technology has been necessary to enable humans to live in most parts of the world. What determines the line?

New comment by aesthesia in "Show HN: Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it"

aesthesia — Fri, 31 Jul 2026 01:23:31 +0000

In those experiments, the effect only happened between models from the same family (and likely the same weight initialization).

New comment by aesthesia in "Agent Skill to Force Docs in ASD-STE100 Simplified Technical English"

aesthesia — Thu, 30 Jul 2026 23:37:20 +0000

How is that ambiguous? The best interpretation I can find where "test" is a verb is an elision: Test [that] B is an alternative to test A. That is an unlikely reading: "test" is a verb in the first instance and a noun in the second, breaking parallelism. And it's just a very unlikely thing to direct someone to do.

New comment by aesthesia in "Advancing the price-performance frontier with GPT‑5.6"

aesthesia — Thu, 30 Jul 2026 19:42:05 +0000

Luna's now cheaper than 5.4-nano (for output tokens). That's a significant improvement.

New comment by aesthesia in "Pacing the frontier"

aesthesia — Wed, 29 Jul 2026 17:18:03 +0000

These aren't assertions being made without arguments or evidence. Maybe you've engaged with those and find them unconvincing. But from what you've said in this thread it seems more like you've dismissed them without serious consideration. My point is not that everyone should just shut up and believe whatever frontier lab employees say, but that they should take a public statement like this as a sign that maybe they should carefully consider these concerns rather than dismiss them out of hand.

New comment by aesthesia in "Truth is not a direction: a Tarski attack on LLM probes"

aesthesia — Wed, 29 Jul 2026 06:16:41 +0000

Quines produce similar issues to self-referential sentences without being directly self-referential. e.g. "'Yields falsehood when preceded by its quotation' yields falsehood when preceded by its quotation."

New comment by aesthesia in "Pacing the frontier"

aesthesia — Wed, 29 Jul 2026 03:31:04 +0000

I mean, Cervantes was parodying medieval romances. That's what Don Quixote was reading, not modern novels.

The point is, it should give you at least some pause when the people most familiar with a technology are among those making the strongest statements about its potential dangers. Extrapolation has worked extremely well in predicting what AI will be capable of over the past 5 years; how confident are you that it will fail before we reach apocalyptic scenarios?

New comment by aesthesia in "Truth is not a direction: a Tarski attack on LLM probes"

aesthesia — Wed, 29 Jul 2026 02:16:06 +0000

Fun, though as hinted at the end, the point of LLM "truth" probes is to measure the model's internal judgment of truthfulness. There's no reason this judgment, even if measured with 100% accuracy, couldn't be mistaken or logically inconsistent.

New comment by aesthesia in "Pacing the frontier"

aesthesia — Wed, 29 Jul 2026 02:04:33 +0000

How many of those concerns came from the people building the technologies? John Carmack wasn't up in arms about how Doom was going to destroy society. Charles Dickens didn't think novels rotted the brain. Johannes Gutenberg wasn't concerned about the effects of a free press.

New comment by aesthesia in "A walk through of the DeltaNet family of linear attention variants"

aesthesia — Tue, 28 Jul 2026 20:20:02 +0000

I think the origin of the title formula is Timothy Chow's "You Could Have Invented Spectral Sequences" (https://www.ams.org/notices/200601/fea-chow.pdf), which is certainly not aimed at an average reader.

New comment by aesthesia in "Our position on open-weights models"

aesthesia — Tue, 28 Jul 2026 00:16:39 +0000

Who pays to build the open weight models? The cost of training a frontier model is orders of magnitude greater than the cost of safety tests.

New comment by aesthesia in "Claude Opus 5"

aesthesia — Sat, 25 Jul 2026 00:40:41 +0000

Yeah, there's probably not a lot systematic behind Anthropic's version numbers. Opus 4.5 was a third of the price of Opus 4.1, indicating there was probably a change in underlying architecture. Opus 4.7 changed tokenizers, probably another base model change.

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

aesthesia — Fri, 24 Jul 2026 23:48:58 +0000

Unless you believe that OpenAI was trying to get their models to break out of the sandbox and hack into Hugging Face, and aiding them in that goal, I don't see how any of those questions would really contradict that claim. The models did something in the real world! It required some amount of complex orchestration and long-term planning!

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

aesthesia — Fri, 24 Jul 2026 23:35:24 +0000

Could there, conceivably, be some third factor at play here? One that both led OpenAI to have concerns about releasing GPT-2 as well as led Microsoft to invest in OpenAI?

New comment by aesthesia in "Nvidia, Microsoft, Meta warn against overregulating open-weight models"

aesthesia — Fri, 24 Jul 2026 19:25:24 +0000

They could start by releasing the MAI-1 models they announced recently. https://microsoft.ai/models/

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

aesthesia — Fri, 24 Jul 2026 19:21:57 +0000

You ask good questions, and I would like to know the answers to them, but I don't think any likely answers would actually change the conclusion much. Any probable set of circumstances that produces this report from OpenAI represents a significant failure on their part to align and control the model.