My ancestors survived by ignoring things like cockroaches so now I gotta get the heebie jeebies every time I see one of these fuckers. Extremely unfair.

Brb making sweeping assessments of the internal states of Google, Anthropic, and OpenAI on the basis of turnover in a high turnover industry.

So, if someone accepts that a skilled user is and will likely remain* necessary to get real use of an LLM in a specific domain, I do not fully understand what the actual pickle is with educating students and so forth. (*The most bullish people probably don't, so this is not relevant to them.)

"We find that humans enrich logical reasoning through pragmatic inferences across languages. Model behavior is more variable." arxiv.org/abs/2605.21299

Is the US preparing a shakedown or did somebody actually managed to smuggle a long truck-sized machine to China? Who knows! www.bloomberg.com/news/article...

Sharing without comment. Looks interesting link.springer.com/article/10.1...

By far, my favorite philosopher is Hofstadter

Interesting!

In fairness to the various Finance Bros, it seems true that, under enough pressure about a looming recession (or worse), the Trump admin. agreed to basically whatever Iran was demanding. On the flipside, I think what's happened since Feb. is not being priced-in properly.

So is this Iran deal actually happening or what

Architect of the Trump administration's AI Action Plan is joining OpenAI.

Didn't Google lock this guy into a contract worth like $2 billion?

Ok foreignpolicy.com/2026/06/17/t...

Re: the idea that leaders at AI labs - or really just Anthropic - believe what they say about RSI, ASI, and so forth. I've found it useful to think about this sort of thing as, yes, they believe what they say, but sincere belief and self-interest can get tied up with one another seamlessly.

An urge to write but no clue to what write about. Awful.

Getting reports that people are doing things with large language models

Preview, not Fable www.nextgov.com/cybersecurit...

I'm building an AI-powered service that alerts you when new AI-powered products you might like become available. Powered by AI.

Sharing not because I'm in tech but because idk what the hell some people in tech are thinking right now.

Currently hearing things about vagueposting 👀

I think the best indirect argument suggesting that LLMs do not do something akin to human reasoning (which is not the same as saying they don't reason at all) is that their writing is not good.

Interesting read. Not sure Carlini is the guy you'd wanna send to ease the government's concerns though. www.wsj.com/tech/ai/anth...

Doesn't make it remotely better, but this is from 2025.

- AI company claims later models will have capability X. - People argue over whether later models will have X. - Company releases model, claims it has X. - Questionable that it has X. - Some people believe it has X, others do not. - Time passes, company claims later models will have Y.

Anybody got any book recs? If we're moots then we're probably interested in similar things anyway, so feel free to just name whatever you might be reading atm.

This spiritually contradicts the Semafor reporting that the reason for the directive was because a China-linked group had *already* accessed Fable/Mythos. www.reuters.com/technology/a...

France is pretty serious about both decoupling the tech it uses for government operations from US firms while also developing a sort of Maven-esque C2 system without Palantir. www.defensenews.com/global/europ...

People are always coming up to me and saying, "your poasts changed my life." I always tell them the same thing: "I know."

One day they're gonna digitally resurrect Al Pacino and call him Al Pacino

I am very interested in this man's arc www.youtube.com/watch?v=vS-g...

All of Western philosophy consists of a series of DLC expansions to Plato.

I will no longer poast about controversial things. It's time for a change. From now on, I will post only about things that there is consensus on. We can all agree that Searle's Chinese Room Argument correctly points out that Large Language Models don't understand a word they're saying.

If there's anything I'm taking from this breaking containment, it's that I'm living with a much different view of human beings than a lot of you. Which is well and good. But interesting to be aware of it.

Updated version of a paper I enjoyed in its ArXiv form: jlm.ipipan.waw.pl/index.php/JL...

A lot of my AI poasting recently has stemmed from me feeling like we're just all talking about this in the wrong way, and I am mostly unable to express what I mean by that.

Totally unrelated to previous poasting, the media is running with this Iran MoU the same way the first "ceasefire" was announced. That backfired for a couple of months. I genuinely do not think this is a done deal. It just ain't over til it's over.

Alright, can I be real here? A world with no labor and where everyone pursues leisurely interests sounds like a dystopia to me.

Are LLM-generated stories novel? They can have unique characters and cliché plots, or the other way around. A holistic score doesn’t help distinguish the two 😔. Meet GENIE 🧞 – a fine-grained novelty metric that tells you where and why a response is original!

Anthropic just shut access to Fable & Mythos, their most powerful models. I wrote up what happened + the policy and psychodrama behind it (with all the links & receipts) here — the thread below is the TL;DR 🧵 blog.geomblog.org/2026/06/the-...

Dawg www.semafor.com/article/06/1...

You're laughing? I just ignored the Treasury Secretary's urgent attempt to reach me and you're laughing?

Peace deal, etc etc bsky.app/profile/vcar...

Pascal only did the whole wager thing as an excuse to indulge in his crippling gambling addiction.

A comment on the stakes of export controls, a topic that I spent 3 years following following fairly closely. 🧵 The Biden admin.'s slate of export controls on the flow of semiconductors to China in October 2022 was a watershed, and somewhat more significant IMHO than this Trump admin. directive.

I do think this is why usage-based linguistics is weaker with reliance on LLMs than without it.

If I could go off this for a moment: I am comfortable using the term "intelligent" to describe a contemporary LLM, provided - crucially - that one understands this just amounts to fitting a set of behaviors to a definition.

Recursively self-improving drama queens

I'm seeing a lot of this with Fable suspension talk, and I still don't understand the extent of it.

Thinking about becoming a Chinese Room Guy