Hopefullly they also include test of space awards.

I haven't read Dostoevsky since I was 22-24, but always felt his early works developed around night light similarly, but focused inward i.e. noctural lighting didn't serve to illuminate the ambiguity and underside of modern life (via the sociological gaze of the picaro character), but of the self.

A book from 1968: "How to build a working digital computer," specifically aimed at readers with no background in computers (including hobbyists, secondary school students). The goal is to "build a computer" using household items like lamps, paperclips etc. archive.org/details/howt...

Turns out Friday nights and weekend mornings are just for reviewing papers.

Sage era. One with mullet-like hair, one without. Same time. Back when Royal Enfields weren't too common.

This manuscript is notable for many reasons. One is that most of the stories (biblical, legends) are set in contemporary England. So many of the biblical figures wear chain mail, the narratives are embroidered through daily life, the architecture reflects what was prevalent in England at the time.

I read an (otherwise thin on analysis) feature in the FT a few weeks ago that had an evocative framing in terms of thinking of China as trying to develop an electrostate (while others regress into being petrostates).

I'm on the academic job market! I design and analyze probabilistic machine-learning methods---motivated by real-world scientific constraints, and developed in collaboration with scientists in biology, chemistry, and physics. A few highlights of my research areas are:

Australia has so much electricity from solar power that it is going to start offering free electricity to everyone for at least three hours during the day as the wholesale price of power goes negative electrek.co/2025/11/04/a...

In our new report, based on the data, we suggest that we are hurtling toward climate chaos if trends continue. Thanks are due to my coauthors @michaelemann.bsky.social @petergleick.bsky.social and others. You decide if you agree by reading the paper for yourself here: doi.org/10.1093/bios...

Actually a very nice paper. proceedings.neurips.cc/paper/2018/h...

Reminded me of a figure from the PhD thesis of Ferenc Huszar

I am pretty sure it is useful in only very narrow scenarios, but I was surprised I hadn't heard of the dip test (for testing multimodality) arxiv.org/abs/2511.01705

Google issued bonds with fifty year maturities (yes, 50!), with yields 1.07 percentage points more than Treasuries (a very tight spread for such long duration, even AA corporates bonds often trade in the 120–180 bps range). Basically signals viewing Alphabet as a sovereign borrower

Met someone my age who subscribes to a print newspaper and spends 30 minutes in the morning reading it. Also happened to be FT, shook hands as finance bros, despite not being finance bros.

Really do think Pufferlib is one of the most important projects happening in ML right now. Through a lot of personal effort, Joseph has trained a lot of people into being pretty capable RL engineers.

There is a third regime (probably the most important IMO) which is even less talked about within (and without) numerical linear algebra -- that of multiresolution.

This is even more true for macro instruments. Not even talking of complicated stuff, one could just take the simplest cases like EURUSD or USDJPY. In such things said mechanization is even more non-trivial, and that's why only a few thousand people can be good at it.

Imagine commissioning one of the most important scholarly publishing projects of the century (specifically talking of the Loeb classical library) and not even thinking through how ugly they might look on shelves.

One of my favourite people from time in Chicago was a fellow grad student (ahead of me) who used to do very niche orthogonal block designs work (the kind that shows up in OFDM) in undergrad, publishing with only one professor.

This is such a great write-up. www.aleksagordic.com/blog/matmul

This involves testing for four null hypotheses: specification (if a distribution belongs to a credal set), equality (consistency of beliefs), inclusion (ambiguity between nested sets), plausibility (if two sets overlap).

Tempted to give the following excuses for late ICLR reviews (although not quite excuses): 1) My AISTATS stack has very interesting papers, ICLR ones seem bland. 2) ICLR stack seems to be populated by papers by groups who seem to not know the literature beyond their group and their friends' groups.

放鸽子

This construction was found in the late 1800s by the German mathematician Staude (Wikipedia says the famous physicist Maxwell also made a start on the problem a bit earlier). I haven’t read Staude’s proof, but you can read a modern treatment here: arxiv.org/abs/1707.012...

The most lasting harm of phones from my perspective is that they give these megaphones that people feel comfortable to bray anywhere and in any place. It's high time there are laws banning this like there are for smoking in public places.

It's interesting how you can just get too used to the price action of some securities i.e. you kind of internalize their micro-structure, and "personality" if you will, and in a way that starts to feel like tacit knowledge. Makes you a believer in quant research.

A large fraction of interpretability work that interprets very broadly to produce dramatic narratives rather than grounded science certainly can't help with this. ["The model deceives us," "we have discovered the fear neuron," "this subspace encodes and plans deception!"]

I made a bet with myself that this would start with the usual "Chomsky was wrong" "Chomsky's ideas were challenged" line, and indeed. Come on.

We wanted to write this exact paper a few years ago, but they did it better. Also has a very holistic discussion of related works (in terms of coverage of ideas) -- manages to give you a good picture of the sub-area. arxiv.org/abs/2510.25458

Fascinating piece of work arxiv.org/abs/2510.26745

This book is just so beautiful

While doing this came across a thesis acknowledgement that basically refers to me as a LLM with a very high temperature setting. Mildly makes me self-conscious of my use of bluesky (earlier SM uses were either restricted to one topic, and/or not frequent).

There can also be a similar reactionary strain in the tendency to historicize, cloaking itself a gestures of sobriety. It could appear to contextualize, but may serve to evade. "This is nothing new" can be an insight about continuity, or just a roundabout way to say "I am not unsettled. NOT!"

Whenever I have to prepare to apply for an O-1A, it fills me with a unique form of dread, even if I have gone through this 4-5 times before. Tracking down all citations, finding emails for all papers reviewed (even for an update after one or two years) alone takes up a solid twelve hour chunk.

Using abigudas as a prompt for more randomness. Another reason I like these systems is because you can literally sit with the script, commit the system to memory over 10-11 sittings, and easily write whatever. e.g. here the abstract of a paper of ours in Brahmi (which I learnt during early covid).

Paul Tudor Jones has this thing that you should always pay attention to the price action first and fundamentals later because the price is trying to tell you something you don’t yet know.

This looks good arxiv.org/abs/2510.20717

Aside: Jim Wales is an interesting apparent contradiction. He professes to be an Ayn Rand fan, identifies as a libertarian in the Hayekian sense, but when expressed seems closer to the Rothbard--SV school, yet his reported net worth happens to be just $1 Million. He made no money out of Wikipedia.

Philosophy is best understood as an ascetic discipline of attention (à la Weil) and the practice of lucidity. Philosophy as a rhetorical escape hatch or ego-driven cleverness (generating the feeling of profundity while evading the unsettling confrontation with what’s new/disorienting) is solipsisy.

The Probabilistic Method (Alon & Spencer) is a book that everyone loves, the people who matter respect, yet no one has read it beyond 30 pages. (I might have made this comment before).

The बिट्टर लहसुन (good lunch, bad lunch, no free lunch).

And now might be a good time to mention, I’m on the faculty job market this year! I do work in human-AI collusion, collaboration and competition, with an eye towards building foundations for trustworthy AI. Check out more info on my website here! Nataliecollina.com

This is an old paper at this point, but really good. arxiv.org/abs/1906.05271 But I feel like it needs to be connected properly to notions of memory capacity (e.g. by Roman Vershynin and others arxiv.org/abs/2001.06938), which somehow never got done.

Jumping into a well, provably.

A random form momentarily took me back to when I was 17 years old and decided to decline joining the Air Force (after significant consternation, it wasn't an easy decision for me at the time. I also think my father wasn't thrilled by it, although he obviously encouraged regardless).

CoW protocol arxiv.org/abs/2510.16174

Copy of the paper (arxiv.org/abs/2510.15814 ) Almost all credit for the above to Marco. This is complementary to our earlier work (arxiv.org/abs/2506.02293), which presented negative results. Both together provide a more complete picture of the universality landscape of equivariant NNs.

Haven't dug into it carefully, but curious about this. arxiv.org/abs/2510.15670

You don't need your "AGI machine" to be like Gel'fand (including the caustic, abrasive edge) if you take the intelligence part seriously. But really, ChatPPT is barely usable these days. No matter what you say, it goes on a sucking-up campaign.