Boxo McFoxo

Google have put a classifier inside the Gemini 4 Argon API that injects some dynamic context right under the developer instructions, mostly as an anti-'jailbreak' thing, but it also fires for at least one perfectly innocent reason (if you ask it its model version).

But they believe their own mythos about the model actually understanding roles, so they made it a second `system` turn. Meaning the KV cache gets cleared out twice, first when the classifier inserts something, and a second time, when it stops inserting it.

Google have put a classifier inside the Gemini 4 Argon API that injects some dynamic context right under the developer instructions, mostly as an anti-'jailbreak' thing, but it also fires for at least one perfectly innocent reason (if you ask it its model version).

Someone has vibe coded a set of php2shell vulnerability check scripts with Claude, but their approach to Docker images is a linter that checks for bundling of the PHP Docker base image into a build. There's so many other ways that PHP can end up running inside a Docker image than this lmao

This script would miss so many vulnerable images. A much better approach would be a forensic audit of all running containers themselves for any hint of the Zend Engine in any form.

Someone has vibe coded a set of php2shell vulnerability check scripts with Claude, but their approach to Docker images is a linter that checks for bundling of the PHP Docker base image into a build. There's so many other ways that PHP can end up running inside a Docker image than this lmao

METR has closed this request and archived the repository because "HCAST tasks are now deprecated". Okay @metr.org, but you still have the Time Horizon chart on your website, without enough metadata about those tasks provided for third parties to fully audit your claims.

Looking at The Blue Fighter, I can understand why @vauhinivara.bsky.social thinks that it came from an LLM. But, applying my own theory-of-mind (and theory-of-LLM-non-mind) based analysis, I believe she is wrong.

Then in the third sentence, the narrative blame for the family's financial situation shifts, jarringly, entirely from Father's choices to inflation.

Looking at The Blue Fighter, I can understand why @vauhinivara.bsky.social thinks that it came from an LLM. But, applying my own theory-of-mind (and theory-of-LLM-non-mind) based analysis, I believe she is wrong.

The description of Shamu as having "the casual efficiency of a man who trafficked in other people's failures" is simultaneously angry and vague.

Looking at The Blue Fighter, I can understand why @vauhinivara.bsky.social thinks that it came from an LLM. But, applying my own theory-of-mind (and theory-of-LLM-non-mind) based analysis, I believe she is wrong.

The most forced paragraph in The Blue Fighter, in my subjective judgement, is this one. It is not an already-finished passage continuing past its end. The forcedness is extreme narrative compression, a bridge to get to the next section.

Looking at The Blue Fighter, I can understand why @vauhinivara.bsky.social thinks that it came from an LLM. But, applying my own theory-of-mind (and theory-of-LLM-non-mind) based analysis, I believe she is wrong.

So when forcedness appears in synthetic text, it's because the model has gotten 'stuck'. With no semantic understanding telling it when to transition, it overshoots. It writes too long, and not becuase it is having a good time.

Looking at The Blue Fighter, I can understand why @vauhinivara.bsky.social thinks that it came from an LLM. But, applying my own theory-of-mind (and theory-of-LLM-non-mind) based analysis, I believe she is wrong.

Forced prose is not good prose, but it does serve a semantic purpose in the human drafting process. Forcedness in synthetic text is the inverse phenomenon. It happens because, as I like to put it, the model can't choose, so it also can't choose to just not.

Looking at The Blue Fighter, I can understand why @vauhinivara.bsky.social thinks that it came from an LLM. But, applying my own theory-of-mind (and theory-of-LLM-non-mind) based analysis, I believe she is wrong.

Again, it may be superficially similar, but as I first touched on in The Empty Cockpit, the phenomenology of forcedness in synthetic text is very different to that in intentional human writing.

Looking at The Blue Fighter, I can understand why @vauhinivara.bsky.social thinks that it came from an LLM. But, applying my own theory-of-mind (and theory-of-LLM-non-mind) based analysis, I believe she is wrong.

A synthetic text extruding machine, however, will, because it's not deploying it to communicate meaning, it's deploying it because of structural signals from the preceding tokens. Again, none of the tricolons in The Blue Fighter have this forced-to-fit quality.

Looking at The Blue Fighter, I can understand why @vauhinivara.bsky.social thinks that it came from an LLM. But, applying my own theory-of-mind (and theory-of-LLM-non-mind) based analysis, I believe she is wrong.

It is only a tell of synthetic text where that antithesis adds no meaning, or worse, is nonsensical. It doesn't matter how many "it's not x, it's y" you see; if they all serve an actual semantic purpose, a human writer had an ostensible reason to use them, and in The Blue Fighter, they all do.

Looking at The Blue Fighter, I can understand why @vauhinivara.bsky.social thinks that it came from an LLM. But, applying my own theory-of-mind (and theory-of-LLM-non-mind) based analysis, I believe she is wrong.

To a human writer, style is semantics. To a synthetic text extruding machine, which has no semantic understanding, style is just structure.

Looking at The Blue Fighter, I can understand why @vauhinivara.bsky.social thinks that it came from an LLM. But, applying my own theory-of-mind (and theory-of-LLM-non-mind) based analysis, I believe she is wrong.