Joshua Grochow

We have *no idea* how to make AI that can do something like that, but we (humans) are evidence that there is a physical process capable of it. And it's pretty reasonable to think that such physical processes shouldn't be restricted to carbon-based neurons.

You don't have to train a child on everything humanity has ever written to get them to speak coherently; you just have to talk to them for a few thousand hours (and let them see you talking to other ppl, maybe correct some mistakes they make, etc).

The training process of current LLMs is *incredibly* inefficient. Compare the energetic costs of training these slop machines to the costs of training a person - over the person's lifetime! Many orders of magnitude worse, and the human is way better at "intelligence", whatever that is.

And many people think that's what needed to get closer to "true AI" is some combo of stats on steroids (~LLMs) and something like "GOFAI" (good-old-fashioned AI), i.e. things like knowledge bases, logical (rather than statistical) inference, etc. "Neurosymbolic" is today's buzzword for that combo.

In an extreme scenario, imagine developing a brain in a vat. It's not quite a person (having a body seems to be an important part of personhood), but I think we'd certainly say it is sentient, intelligent, etc. Like, is Data's "neuromorphic processor" essentially just a manufactured brain?

Sure. I mean, I still believe that in some far-distant future (if we survive climate catastrophe), "true AI" could be possible, essentially b/c I believe that we boil down to physics (even allowing the possibility that some of that physics is not yet completely understood or developed).

Right. (can elaborate if wanted)

I would not classify "ppl who work on AI" as one group. Good researchers (AI or otherwise) know what you say. But many ppl seem to have completely drunk the kool-aid of the current genAI hype and don't seem to understand what you said.

Wait doesn't any nondegenerate bilinear form do this? (at least in the finite dimensional case)

Ok so yeah, this has quickly become the #1 misunderstanding about the canceled grants

the grants are not “subsidies” or “entitlements” to Harvard or Princeton or whatever

they aren’t going into universities’ endowments

they are competitive contracts won by these universities to do research

There is a podcast interview with her! Props to @optima-arc.bsky.social for recording this important history

Applications are open for SFI's 2026 Complexity Postdoctoral Fellowships

If you’ve recently earned a Ph.D. in any scientific field and want to pursue independent, transdisciplinary research, consider applying.

Deadline: October 1, 2025
Apply here: santafe.edu/sfifellowship

Description Please note that job descriptions are not exhaustive, and you may be asked to take on additional duties that align with the key responsibilities ment...

I am advertising for 4 post-docs to come to Imperial and formalize, in Lean, *statements* of theorems from recent issues of the top generalist pure mathematics journals.

www.imperial.ac.uk/jobs/search-...

Positions are for 2 years, start date 1st Oct this year. Deadline 15th August.

I think this is University of Virginia's gardens near the main forum building, showing a red brick wall that is serpentine in shape, casting an odd undulating shadow on the white-paved road/driveway nearby.

The curves (modeled after a sine curve) increase the strength of the single brick wall to toppling by effectively widening the base.

To accomplish the same task with a straight fence, we'd need, at a minimum, two brick width. So the serpentine reduces the brick needed per linear length by ~70%.

In our Theory Group zulip server they added this as a custom emoji in response to something someone (who shall remain nameless) posted

Woohoo! Can't wait!

The First Planned Migration of an Entire Country Is Underway The Pacific island nation of Tuvalu could be submerged in 25 years due to rising sea levels, so a plan is being implemented to relocate its population to Australia.

The first planned migration of an entire country due to climate change is happening! Tuvalu, a Pacific island nation, is at risk of being submerged under rising sea level, so it has signed an agreement with Australia to allow 280 Tuvaluans to settle in Australia as permanent residents each year.

"True basically everywhere" - really? My impression was that it was much more true at e.g. Brown than at most other schools.

On the individual level, if your assessments practices don’t allow for revision and don’t leave space for scaffolding and reflection, you are suggesting that the essay, the narrow tip of the wedge, is all that matters. And students are going to take whatever steps necessary to make sure they nail it

Not just writing!

In math & CS classes, for so long w/ traditional grading the emphasis was on Getting The Right Answer. Gave the impression that the output was the point, rather than the learning. Showing the right answer was just a mediocre attempt to assess whether & what a student had learned.

Ironically, clause 3(a) "truth-seeking" excludes all LLMs

I don’t know who needs to hear this but we don’t need to consume the Greenland ice shelf to organize our email

It's absolutely ridiculous — an embarrassment — that algebraic topologists cannot algorithmically compute Morava K-theory of arbitrary cell complexes. The best they can do is compute the E₃-page of the AHSS and hope that fate smiles on them. Absurd! /rant 🧮

Wait I can't tell if this is fr or not. Is it really not known to be computable? (In the Turing sense, not the sense of "can I actually run this on a computer in the real world and get an answer")

we at @aial.ie, are investigating amplification/censorship on X/twitter in EU. what are major EU:
-politicians & regulators
-journalists
-influencers & public intellectuals, with high visibility/influence on X, across both left & right political ideologies

we're interested in existing datasets too

I would love if you all would give me suggestions for papers about AI well suited for TikTok style explainers! Stuff that’s important or interesting or just good for more people to know about. (It’s ok if it’s your own paper!)

I agree pattern matching is fine, maybe even better than autocomplete for certain audiences.

We apparently disagree about whether or not one can the safely anthropomorphize in that context, and seems like neither of us will convince the other. 🤷

Anyway, if it helps, any time you’re tempted to say something about “AI,” think about whether you’d say it about your toaster. You programmed your toaster for three minutes to make toast and it burned the bread?

Your toaster didn’t “lie” to you. Your toaster is broken.

Starlink's connection speeds drop below the federal definition of "broadband" at a density of ~7 customers per square mile, according to recent research. The more the customers in an area, the slower everybody's connections get. Data shows that 83% of US Starlink customers get sub-broadband speeds.

"pattern matching" as you suggest is fine too, and maybe even better for a tech audience. But notice how far "pattern matching" is from the anthropomorphized "it desperately wants to be nice to you"