Jeffrey M Epstein

Saw a bus stop ad advertising some ai for business thing, and the scenario was a guy with an ice cream store asking “what were our most popular flavors this month?” Have these people not heard of excel?

my first book (2020) was about online neonazis and my second book (2024) was about christian theocrats. they gained strength, voltronned and became a government. my third book (2027) is going to be a history of notable sandwiches so i hope this means we will be governed by sandwichocracy

as a member of the canadian diaspora what aspect of canadian identity should I get weird about

"I use sonnet" "I use haiku" name one poet bro

Got it. I'm sorry for your loss. To clarify: You want a eulogy for your father that's 𝘂𝗻𝗶𝗾𝘂𝗲, sounds 𝘀𝗺𝗮𝗿𝘁, and captures the 𝗹𝗼𝘃𝗲 𝘆𝗼𝘂 𝗵𝗮𝘃𝗲 𝗳𝗼𝗿 𝗵𝗶𝗺. I get it. He wasn't just your father — he was your Dad. Here a few options I came up with:

It would be so cool if we had machines that were really good at doing exactly what we tell them to do when we give them precise instructions. Oh well, might have read too much sci-fi as a kid.

Congressman Brandon Gill @RepBrandonGill How many Americans have to get Allahu Akbar’ed before we realize Islam is a problem?

The way sitting politicians can say this stuff about Muslims while facing zero consequences is an indication of just how entrenched Islamophobia still is in our institutions and culture.

Too many “middle ground” AI arguments—“I have concerns, too, but we have to adapt”—proceed from what is to me a peculiar embrace of “inevitability” which seems to be magical thinking, a way of depoliticizing the political, of self-soothing in the face of an overwhelming challenge.

US-Israeli strike on girls' school kills at least 85 students, Iran's judiciary says At least 85 students have been killed in US-Israeli strikes that hit an Iranian girls' school in Hormozgan province, Iran's judiciary said on Saturday. Washington has not commented on the reported str...

We killed 85 schoolgirls. We are not the good guys.

Reading literature can be like that certainly, as can reading scientific writing. If I don’t understand what an LLM means by some opaque passage, it’s probably because it doesn’t mean anything at all.

In general you can’t! But I meant more that often a piece of text initially seems not to make sense, but you say “well probably the person who wrote this had some consistent thought in mind, so I should try to understand it”. With LLM generated text there was no thought, it might just be nonsense.

It’s been really disappointing to me to realize, maybe belatedly, that a lot of people just do not see the point of not doing something you think is bad if other people are probably going to do that bad thing either way.

Writing "literally fucking anything you can imagine" in my list of future predictions so everyone praises my insight when it becomes true.

Imagine waking up and your government ID has been declared invalid. It’s now illegal for you to drive, or do anything else that requires government ID. You had no warning. You can be arrested if found driving.

That’s life for trans people in Kansas now.

I don’t think most academics realise that we’re way past the ‘AI concerns me’ phase and in the middle of a full scale apocalyptic war waged by tech corporations. We must fight for our lives and for everything we hold dear. AI should be exposed, shamed, and rejected in every facet of our profession.

I think another important aspect of this is that when you engage with a text written by another human, you can rely on there being a thought process more or less like your own behind it, which is often critical for interpretation.

Yeah it’s definitely what he says he shows

I don’t yet see how he can even claim that it’s consistent to do so, given that he also claims to be working entirely within unitary QM. Discard from our consideration, sure, but that’s what Everett does. But discarding from “reality” just has to involve something other than unitary evolution.

That particular article with Preskill I think is less research paper and more two very eminent researchers speculating. Both have a ton of much more detailed, less speculative work.

Oh it's definitely that!

Yes and no. I think these careful but limited works are at least interesting, and on the other hand I do think there's a lot of work, which obviously I won't call out, that is basically just jumping on the bandwagon of combining the two buzzwords, and adds nothing of value.

Based on what I know of his other work, he's a very careful and rigorous scientist, so I feel pretty confident that whatever is in these papers (I haven't carefully read them), it is not just hype or anything like that. But in general I haven't seen much that has convinced me to hope for a lot here.

First they came with the snowballs, and I said nothing, because I'm not a crybaby

And this is in some sense what that first link, about kernel concentration gets at: yes quantum computers work in exponentially large Hilbert spaces, but that doesn't mean they can "handle lots of data".

Absolutely -- the big misconceptions about quantum computers are that they are (will be lol) just faster computers, that they "do lots of computations in parallel and pick the correct one". In fact, only a small number of very structured problems has been shown to admit a quantum advantage.

Finally there's this work (+ more) from Jens Eisert and co (he leads a huge group doing fantastic Q. computing/info work). I heard him talk a few months ago, where he actually expressed a lot of optimism about quantum speedups in ML in general.

arxiv.org/pdf/2502.14252
arxiv.org/pdf/2510.19928

Robert Huang is someone who has done a lot of work on quantum algorithms for dealing with quantum data, which might be a promising direction, though I haven't really looked into his work. (He gives good talks, though.)

scholar.google.com/citations?us...

There is work that demonstrates provable quantum speedups on artificial problems, based on the classical hardness and quantum efficiency of discrete log. This is interesting, but of course probably not terribly practically useful.

arxiv.org/pdf/2010.02174
arxiv.org/pdf/2011.01938

Exponential concentration in quantum kernel methods - Nature Communications Quantum kernel methods are usually believed to enjoy better trainability than quantum neural networks which may suffer from a well-studied barren plateau. Here, building over previous evidence, the au...

On the ML side, there was early hope that interesting kernels could be computed using quantum computers and then used to do classical ML, but there are some no-go theorems that make that seem unlikely, see e.g. this work www.nature.com/articles/s41....

Not all of us get so much investment in validating our work!