Any really good actually insightful or helpful articles or resources on teaching/learning in the age of LLMs?

Some misc observations on trying to use LLMs to help on a current research project…

Can we infer chemical reaction networks from time series data? Can we observe concentrations of substances over time and infer which substances react with each other? Learn if (and how much) this is possible from Yong See Foo, @adrianazanca.bsky.social and JenniferFlegg arxiv.org/abs/2505.15653

Evolutionary algorithms are back maybe?

I tried defining a simple ‘prompt script language’ that sort of works and is fun to play with gist.github.com/omaclaren/b5...

I love you bluesky but I admit I don’t totally understand the rabid and uncritical AI hate that predominates here

The voice mode for all the llms keeps trying to sound more human but I kinda just want my robot friends to sound like robots

I’ve got some students who wanna play around with eg 70B open llms. Can’t run locally obvs. What’re their best options in terms of ease of use and cost? Mainly just running for now but also maybe fine tuning at some point

No matter how much beef you have with someone you don’t need to do this shit. Crediting people appropriately is allowed

Here’s a thing. A bit bitter sweet for a few reasons, and also my name is spelled wrong lol but still www.sciencedirect.com/journal/jour...

Claude Code is very cool and helpful but also

Smashing the endorse button as fast as I can www.lesswrong.com/posts/oKAFFv...

Considering switching to cursor solely because the grammarly desktop widget thing works in it but not vscode

Claude is back 🙏 but also even anthropic is still at this point

This came up recently but I’m not 100% sure how to think about it. Eg suppose you have a nonlinear function of two variables f(x,y) evaluated on a 2D grid. One apparently useful way to approximate it is to think of the values as entries in a matrix and use matrix factorisation. But why?

Idle stats thought — calling them confidence subsets/subintervals instead of confidence sets/intervals would more honestly indicate that the coverage guarantees hold with respect to some set of initial of values considered

Do large language models develop "emergent" models of the world? My latest Substack posts explore this claim and more generally the nature of "world models": LLMs and World Models, Part 1: aiguide.substack.com/p/llms-and-w... LLMs and World Models, Part 2: aiguide.substack.com/p/llms-and-w...

Hello, here is some new work called ‘Invariant Image Reparameterisation: A Unified Approach to Structural and Practical Identifiability and Model Reduction’. See arxiv.org/abs/2502.04867

Shout out to this random book exchange setup I found while taking my daughter for a walk!

A nice summary on the other site of RL self improvement by @ptrschmdtnlsn.bsky.social “we keep distilling what we conclude after a lot of thinking into what we conclude intuitively in a single step of thinking, which in turn improves what we conclude with a lot of thinking, and so on”

Just finished @melaniemitchell.bsky.social ‘s really nice book on AI. Just pre- chatgpt etc but good coverage of context leading up to recent explosion. Based on this quote I’d say at least 25 years to ‘AGI’. Will be interesting to see…

Man still only o1 can solve this. Not Claude, not DeepSeek not Gemini … (DeepSeek app can’t even extract the words from the image correctly…had to pass manually)

We are advertising two PhD scholarships to work on models of the interaction between ethnicity and infectious disease dynamics at the University of Canterbury in New Zealand. Open to international students. Please share! www.findaphd.com/phds/project... @iddjobs.bsky.social

I have been trying to get more familiar with the basics and advances Reinforcement Learning. I wonder if anyone have seen a paperthat use potential outcome notation within the RL world. This would make the life of many economists much easier. Anyone? All help is appreciated!

My ACM Crossroads article is out now! Should be a fun read if you're interested in a broad overview of my computational photography work: dl.acm.org/doi/10.1145/...

Yes exactly

This seems like a pretty reasonable take on the impact of (current) AI tools, applicable beyond software open.substack.com/pub/pragmati...

On the one hand, yes, on the other hand …

Getting these emails including one on Xmas day from the same journal lol … it’s the holidays man leave us alone

Bluesky can be a fraught place to post about AI but it is worth noting that the buzz over o1 (& now o3) is not “hype.” We know o1 can actually do some very hard tasks (see my post) & o3 appears to represent a big further leap. They aren’t AGI, but will matter. www.oneusefulthing.org/p/what-just-...

Kinda feel like this might apply to us too, tho

Genuine lol at ‘You're talking as if there's a better way than Bayesianism. But in truth there's no alternative’. Cool thanks mr philosopher

Liking this interaction with @mmbronstein.bsky.social and Denis Danilov so much I'm reposting it here

Tried Claude and o1 on NY Times connection puzzle with first group provided…regret to inform that Claude struggled while o1 nailed it

Had a student come up to me on the uni bball court and say the second-year course I just taught on ‘engineering machine learning’ was their favourite of the year 😃. Good time to say the notes from my half of the course are here: github.com/omaclaren/op...

Great post, some excerpts

Generating cat videos is nice, but what if you could tackle real scientific problems with the same methods? 🧪🌌 Introducing The Well: 16 datasets (15TB) for Machine Learning, from astrophysics to fluid dynamics and biology. 🐙: github.com/PolymathicAI... 📜: openreview.net/pdf?id=00Sx5...

Since I am suddenly inundated with statistics-oriented followers, I should post again that I am still conducting free weekly open office-hours for anyone in the world with Econometrics questions, Wed 10-12AM Eastern or by appointment. Details and sign up at donskerclass.github.io/OfficeHours....

Our JSS article is out! And now I get to focus on {marginaleffects} 1.0.0. Stay tuned. www.jstatsoft.org/article/view...

I had zero luck with grants. People really didn’t get why what I did was important and never figured out how to explain it to them.

I still don’t fully grasp how you can have property 2 without implying 1, as is indicated to be possible in this paper (first two screenshots) Similar idea indicated in Wasserman’s book (third)

The "standard" rearrangement inequality states that is (aₖ)ₖ, (bₖ)ₖ are two finite sequences of n numbers, the sum ∑ₖaₖbₖ is maximized when they are both in non-decreasing order. Very natural! Did you know there was an analogue for functions and integrals? Enters the Hardy–Littlewood inequality:

🥰

Confusing and trivial, but fun and instructive for students. The variance of a random variable X with mean μ is: Var(X)=𝔼[(X−μ)(X−c)] where c is an arbitrary constant

Only tangentially related but I personally like to think of even standard intervention as ‘do then view’ where the ‘do’ is stochastic (remove the arrow, keep the marginal), and ‘view’ is just ordinary conditional probability

"There’s no way you can just sit down & do a `big thing', or at least I can’t. So I just went back to doing lots of little things, & hoping that some of them will turn out okay. Statistics is a wonderfully forgiving field... all you have to do is get an idea & keep at it." - Brad Efron #statsquotes

Whooping cough vaccination rates 'dreadful' in Aotearoa, expert says www.rnz.co.nz/news/nationa...

Since I've been here for a while… here are some useful features on Bluesky that you might not have discovered yet! 🧵

How it started: "AI is going to replace randomized clinical trials" How it is going: ai.nejm.org/doi/10.1056/...