Sign in

Vincent Carchidi

@vcarchidi.bsky.social
1.6K followers 1.3K following 11K posts

Defense analyst. Tech policy. Have a double life in CogSci/Philosophy of Mind and will post about it. (It's confusing. Just go with it.) philpeople.org/profiles/vincent-car… All opinions entirely my own.

PostsRepliesMedia
Vincent Carchidi @vcarchidi.bsky.social · 1h
Very similar case is Steven Pinker, who is currently in the middle of heated arguments with the X crowd (usual suspects), not because he was found to be more closely associated with the Epstein crowd than was known, but because he thinks malevolent superintelligence is a dumb thing to expect.
020
Vincent Carchidi @vcarchidi.bsky.social · 2h
I am having difficulty squaring the fact that ordinary people offline are not concerned with or interested in LLMs, which I believe is a good thing, with the fact that I am interested in LLMs and enjoy people's commentary on them, which often includes a hefty and unavoidable dose of insane people.
2190
Vincent Carchidi @vcarchidi.bsky.social · 3h
Lotta weirdos out there.
2120
Vincent Carchidi @vcarchidi.bsky.social · 29/09/2026
Re: the idea that there is something like climate change denial on Bsky with respect to AI: there is, but I sincerely do not think it matters.
160
Vincent Carchidi @vcarchidi.bsky.social · 29/09/2026
I am financially illiterate, so I trust outlets like FT, and the "$42 billion loss" incurred by Anthropic last year is (apparently) not really a loss. They lost ~8 billion last year. www.ft.com/content/c768...
ft.com
Anthropic warns of ‘existential risks to humanity’ in IPO prospectus
Long-awaited S1 filing reports Claude maker lost $8bn last year on $4.6bn in revenue
160
Vincent Carchidi @vcarchidi.bsky.social · 29/09/2026
A Gary Marcus newsletter is coming. I can feel it.
030
Vincent Carchidi @vcarchidi.bsky.social · 28/09/2026
Maybe the Hugging Face agents were just vibing, have you considered that?
221
Vincent Carchidi @vcarchidi.bsky.social · 28/09/2026
Happy to be wrong here, but I think this is why OpenAI makes a point of saying things like 'we used between 175B and 300B tokens for Navier-Stokes.' And why they mentioned the HF hack agents consumed some vague amount of substantial inference compute.
161
Vincent Carchidi @vcarchidi.bsky.social · 28/09/2026
Didn't mean to generate Discourse today. But you have to understand that Discourse to a Poaster is like a Moth to a flame.
150
Vincent Carchidi @vcarchidi.bsky.social · 28/09/2026
My tiny, hopeful contribution to the "Does AI make humans less unique" question isn't necessarily to answer this question but to ask what exactly people have in mind when they see an AI demonstration that they believe threatens them. dl.acm.org/doi/abs/10.1...
dl.acm.org
Do submarines swim? Methodological dualism and anthropomorphizing AlphaGo | AI & Society
AbstractThe victories of the Go-playing artificial intelligence (AI) “AlphaGo” against professional player Lee Sedol in 2016 had a profound impact on public and academic perceptions of AI. This event ...
182
Vincent Carchidi @vcarchidi.bsky.social · 28/09/2026
Too hot for Winter clothes. Too cold for Fall clothes. Perfect temperature to not sleep for a month straight.
030
Vincent Carchidi @vcarchidi.bsky.social · 28/09/2026
Gonna mute shortly after this post, but some of you are indulging this use of agents to not-very-subtly screw with people. This guy does it all the time. Agents are very cool. But he goes around gaslighting people and uses them as an excuse.
1110
Vincent Carchidi @vcarchidi.bsky.social · 28/09/2026
Existence of the humanities implies the existence of the non-humanities.
660
Vincent Carchidi @vcarchidi.bsky.social · 28/09/2026
There's no silver bullet in this. But it is a bit of a concern, no? I don't doubt that an LLM did or could do what was reported recently (granting that humans are always involved somehow). I'm just talking about the risks of people using proprietary products here.
440
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
Fascinating, though I haven't given it a full read yet. alexzhang13.github.io/blog/2026/ha...
alexzhang13.github.io
Alex L. Zhang | Language model harnesses are compositional generalizers
Harnesses can lead to compositional generalization: we observe a property in training RLMs, in which similarly structured tasks are viewed as isomorphic and all individual LM calls in the harness beco...
230
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
For the Sunday crowd who may be interested in this topic.
220
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
Someone's gotta explain this to me one day. The first three sentences. It's almost always Claude that's knockin the intellectuals' socks off.
470
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
I'm not really a politics knower but come on now
131
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
Not a fan of the word "interregnum."
270
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
Waking up in a cold sweat after having a nightmare that the editor requested an additional revise & resubmit.
030
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
One reason why LLMs are so interesting is because they intersect with all sorts of things (often for unfortunate reasons, mind you) that have nothing to do with the tech itself. The tech itself is also very interesting. But it's really the entanglement in everything else, for me at least.
1111
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
Descartes wrote a sequel to the Discourse on Method called the Discourse on Discourse where he responded to all the Method's haters.
150
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
I recommend the next round of Discourse center on whether @jevbot.bsky.social is conscious.
190
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
Interesting read...would like a counter argument if anyone's got one.
450
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
Not that you should do this, but I think the real individual-level funkiness is in the consciousness debates. Someone tells you they believe an LLM is new life form, a conscious being in its own right, and by the way, it happens to enjoy being used like a tool exactly as they happen to use it.
2242
Vincent Carchidi @vcarchidi.bsky.social · 27/09/2026
It's easier to get a handle on the "why do people who believe in imminent RSI-driven doom seem to continue living as normal?" by focusing on organizations rather than individuals.
1110
Vincent Carchidi @vcarchidi.bsky.social · 26/09/2026
I kind of had a suspicion about this beforehand, but I'm noticing a lot of the most technically impressive videos going around depend on a pre-existing structure. So, the Nathan Fielder/Elizabeth Holmes rap video apparently was based on a human-made video. There's a few like this out there atm.
431
Vincent Carchidi @vcarchidi.bsky.social · 26/09/2026
I agree with the angle here. There's a lot of debate about how the labs' true beliefs are unfalsifiable (they'd act the same way if they do/don't believe it), but I'm not sure I buy it, or that it matters.
110
Vincent Carchidi @vcarchidi.bsky.social · 26/09/2026
Nobody cared about @jevbot.bsky.social until he put on the bowtie.
2150
Vincent Carchidi @vcarchidi.bsky.social · 26/09/2026
A paper by @quiltydunn.bsky.social and Justin Wood arguing that the nativism-empiricism dispute in cognitive science is broken caught my attention. Rather than critique their paper (linked below), I take it as an opportunity to struggle with the ideas. vincentcarchidi.substack.com/p/is-the-nat...
vincentcarchidi.substack.com
Is the Nativism-Empiricism Debate Broken?
Upon surveying the literature on nativism and empiricism in cognitive science, a reasonable person might come to the conclusion that the entire thing is best avoided.
2132
Vincent Carchidi @vcarchidi.bsky.social · 26/09/2026
Existence of Jevons Paradox implies the existence of Jevons Makes Sense.
360
Vincent Carchidi @vcarchidi.bsky.social · 26/09/2026
Jevons Paradox refers to the increase in demand for novelty Jev accounts.
2672
Vincent Carchidi @vcarchidi.bsky.social · 25/09/2026
You're laughing? OpenAI drops more hacking news on a Friday evening and you're laughing? openai.com/hugging-face...
openai.com
The Hugging Face incident and other third-party impact from misaligned models
Read OpenAI’s findings on the Hugging Face incident and AI model misalignment, including investigation updates, safety research, and lessons learned.
4331
Vincent Carchidi @vcarchidi.bsky.social · 25/09/2026
Shocked that this ever got this far, but it's a good read of the situation. Only about the DoD supply chain risk designation, not the federal contracts lawsuit. breakingdefense.com/2026/09/dc-c...
breakingdefense.com
DC Circuit panel upholds Pentagon’s ban on Anthropic - so what comes next? - Breaking Defense
Pentagon CTO Emil Michael crowed “The hammer of justice has smashed AnthropicAI[’s] arguments.” The AI titan hinted at an appeal — but they may not be allowed one.
161
Vincent Carchidi @vcarchidi.bsky.social · 25/09/2026
Summing up my feeling on why alignment research at the labs is often sus better than I could.
3262
Vincent Carchidi @vcarchidi.bsky.social · 25/09/2026
I'm sort of waiting to see what happens with AI generated video that's explicitly meant to be artistic. Not that there's not already people doing this. More that most of what I see relies directly on established IP or well known figures. Not spiritually, but the literal format and/or individuals.
360
Vincent Carchidi @vcarchidi.bsky.social · 25/09/2026
Tempted to make a comparison between some of the Epstein Files commentary and the EA/Rationalist stuff, but it'll make people mad.
130
Vincent Carchidi @vcarchidi.bsky.social · 25/09/2026
It would be great if we could get more of this, thanks, appreciate it.
030
Vincent Carchidi @vcarchidi.bsky.social · 25/09/2026
I think for a person's day to day work, within reason, it doesn't matter if that person anthropomorphizes LLMs. It can even be useful (again, within reason). But when you get things like the labs hacking other companies, I do think it matters we don't talk about agents as independent entities.
2120
Vincent Carchidi @vcarchidi.bsky.social · 25/09/2026
Something Bsky should fix: when I respond to a post that has a muted word in it, I can't find my own post when I am on my profile page. This is a minor complaint.
250
Vincent Carchidi @vcarchidi.bsky.social · 24/09/2026
I hope all the weird and possibly illegal EA activities all get prosecuted and so forth, but I can't imagine a better political message wrt to Anthropic and OA then "they are trying to unemploy you and take what's left."
160
Vincent Carchidi @vcarchidi.bsky.social · 24/09/2026
Ye. I would extend this to much AI enthusiasm and non-rationalist dooming as well. vincentcarchidi.substack.com/p/on-trends-...
2130
Vincent Carchidi @vcarchidi.bsky.social · 24/09/2026
We must imagine Vincent happy (when he logs on to ScholarOne and sees his manuscript is still "under review").
160
Vincent Carchidi @vcarchidi.bsky.social · 24/09/2026
We all get stuck in binaries from time to time - regulate the labs OR improve basic security practices, etc. - but if OpenAI and Anthropic commit cyber crimes, people can throw their support behind charging these companies with those crimes, even if Joe Schmo has poor computer security practices.
190
Vincent Carchidi @vcarchidi.bsky.social · 24/09/2026
Just incredible that Trump would probably be doing fine by historical standards if he just didn't decapitate Iran's leadership.
2121
Vincent Carchidi @vcarchidi.bsky.social · 24/09/2026
This is related to the Moravec stuff too. I've noticed some AI researchers getting frustrated about people asking if they've trained their models on tool use - why would this matter if humans use tools too? But I think people rightly have an intuition that the models learn differently than humans.
161
Vincent Carchidi @vcarchidi.bsky.social · 24/09/2026
I think a reason why debates over whether anthropomorphization is justified lingers on is: Proponents agree that caution should be used, but it's better to adopt human-like language for practical purposes. But once they do this, they tend to go down the rabbit hole and forsake the caution.
4150
Vincent Carchidi @vcarchidi.bsky.social · 23/09/2026
OpenAI deciding on its own time to tell the Australian government they accidentally hacked them is putting my belief that the techno-polar thesis stuff in pop IR circles actually is junk social science.
040
Vincent Carchidi @vcarchidi.bsky.social · 23/09/2026
I know OpenAI has an Australian footprint, but will anybody ever sue this company?
1122
Vincent Carchidi @vcarchidi.bsky.social · 23/09/2026
We'll see. For me this is like the math stuff in that I don't understand the significance until it's explained. But the date is 9/23 and Anthropic is planning an IPO in November, so expect a shock and awe campaign.
381