Hacker News: x312

New comment by x312 in "GPT-6 Astra"

x312 — Thu, 03 Sep 2026 19:50:24 +0000

Hmm, 61 on ArtificialAnalysis, effectively matching GPT-5.6 and trailing the new Meta model. How is that possible along with the other metrics they shared? Insanely jagged intelligence?

New comment by x312 in "Ox Alpha"

x312 — Fri, 21 Aug 2026 02:09:04 +0000

I believe its the same as free models in general on Openrouter, 1k requests per day for accounts that have some spend history.

Misaligned AIs could use killer robots to take over

x312 — Tue, 11 Aug 2026 19:44:19 +0000

Article URL: https://www.lesswrong.com/posts/9jKhqmFjMzdAvHANr/misaligned-ais-could-use-killer-robots-to-take-over

Comments URL: https://news.ycombinator.com/item?id=49263437

Points: 7

# Comments: 4

New comment by x312 in "Stealing Reasoning Traces from Proprietary LLM APIs"

x312 — Tue, 11 Aug 2026 15:16:33 +0000

The provider decrypts it and puts the decrypted reasoning into the model's context window. They prompt the model to repeat back the reasoning. So then the model echoes it back in plain text.

New comment by x312 in "Stealing Reasoning Traces from Proprietary LLM APIs"

x312 — Tue, 11 Aug 2026 15:14:03 +0000

Super cool that this works. I'm surprised these companies re-use the same encryption key across models!

I wonder if you can use these for attacks, like this previous paper showing that if you know how a model reasons, you can "fake its thinking" to control it? https://news.ycombinator.com/item?id=48631888

Existential Risk from AI: An Exposition for Mathematicians

x312 — Mon, 03 Aug 2026 21:41:26 +0000

Article URL: https://alkjash.github.io/ai-risk/

Comments URL: https://news.ycombinator.com/item?id=49161830

Points: 2

# Comments: 0

New comment by x312 in "DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis"

x312 — Fri, 31 Jul 2026 18:15:32 +0000

Have you tried Novita? They're zdr and I find they usually have better cache hit rates than fireworks. No affiliation.

New comment by x312 in "Coding agents think ahead of time"

x312 — Tue, 14 Jul 2026 13:22:00 +0000

It's been known for several years that LLM activations encode future tokens ahead of time (e.g. https://arxiv.org/abs/2404.00859).

But this has only been shown on simple tasks, so I think this paper is still quite neat. The interesting thing is that they show "future horizon length" varies across models.

New comment by x312 in "Let's talk about: LinkedIn ghost jobs"

x312 — Sun, 12 Jul 2026 20:48:05 +0000

Considering the job description is unrelated to the company and the title, this is a scam post, not quite a "ghost job".

They're probably rapidly opening + closing new jobs to increase visibility, as matching models on job boards tend to prioritize new posts.

New comment by x312 in "Grok 4.5"

x312 — Wed, 08 Jul 2026 18:25:44 +0000

Given their pricing, I'd guess their models are just way bigger in parameter count. They've always underperformed in cost-per-performance.

They also target a cost-insensitive market (corporate/coding users) compared to Google/OpenAI which support massive amounts of free users.

New comment by x312 in "Plotnine"

x312 — Tue, 23 Jun 2026 14:07:55 +0000

Love your work on this, thanks for bringing the ggplot syntax to Python!

New comment by x312 in "Prompt Injection as Role Confusion"

x312 — Tue, 23 Jun 2026 01:32:07 +0000

Yeah, the footnote/sidenote on the paper (the one labeled #2) mentions this as well so you can't type that directly

New comment by x312 in "Prompt Injection as Role Confusion"

x312 — Mon, 22 Jun 2026 20:01:13 +0000

I believe they are trained for security now, but you're not wrong in that it's kind of stapled on top

https://arxiv.org/abs/2404.13208

Prompt Injection as Role Confusion

x312 — Mon, 22 Jun 2026 15:48:55 +0000

https://arxiv.org/abs/2603.12277

Comments URL: https://news.ycombinator.com/item?id=48631888

Points: 235

# Comments: 116

New comment by x312 in "GLM 5.2 vs. Opus"

x312 — Mon, 22 Jun 2026 15:37:33 +0000

A lot of open weight models don't understand intent well, they'll overfixate on a word in the prompt or just go off the rails trying to do much work.

GLM-5.2 actually has really good intent understanding though, on par with GPT-5.5 and Opus from my experience.

New comment by x312 in "Rio de Janeiro's "homegrown" LLM appears to be a merge of an existing model"

x312 — Sun, 14 Jun 2026 18:49:39 +0000

This works because Nex itself is a finetune of Qwen3.5 (https://huggingface.co/nex-agi/Nex-N2-Pro). It's merging Qwen3.5 with a Qwen3.5 finetune.

I don't believe this would work on two LLMs that have different pretraining. Even if it did you would need two LLMs that have exact same internal activation shapes, dimensions, expert counts, token vocabulary, realistically it would never happen outside of finetunes or academic experiments.

Rethinking the Value of Arbitrary Order in Diffusion Models

x312 — Wed, 10 Jun 2026 19:00:00 +0000

Article URL: https://arxiv.org/abs/2601.15165

Comments URL: https://news.ycombinator.com/item?id=48481037

Points: 2

# Comments: 0

New comment by x312 in "Where is the AI jobs crisis?"

x312 — Tue, 09 Jun 2026 21:24:15 +0000

True, but coding agents have only been practically useful since early 2025. Jobs are flat or a bit up since then.

New comment by x312 in "Natural Language Autoencoders: Turning Claude's Thoughts into Text"

x312 — Thu, 07 May 2026 22:13:50 +0000

This paper has an major issue that they are not surfacing, these activations can just be correlated on a common latent. For example, both the original activation and the explanation could share a broad latent like "this is an adversarial scenario". That could make reconstruction loss look good without showing that the actual explanation was the correct cause for the LLM's response.

I find this rather disturbing. Anthropic has quite a habit of overclaiming on questionable research results when they definitely know better. For example, their linked circuits blogpost ("The Biology of LLMs") was released after these methods were known to have major credibility issues in the field (e.g., see this from Deepmind - https://www.lesswrong.com/posts/4uXCAJNuPKtKBsi28/negative-r...). Similarly this new blog is heavily based on another academic paper (LatentQA) and the correlation/causation issue is already known.

Shoddy methodology is whatever, but it feels like this is always been done intentionally with the goal of trying to humanize LLMs or overhype their similarities to biological entities. What is the agenda here?