Lightnews — Scholar-powered news

Marcelo Rinesi @marcelorinesi.bsky.social · 4h

#RisingOnArxiv (ignoring a lot of genAI stuff):

* ICCV 2025
* Fermion sign problem
* Genuine multipartite entanglement

Rising on arXiv - 2025-10-05

A few of the ideas, topics, and commonplaces that have been gaining steam on arXiv during the last few months (yes, I do a lot of filtering to keep away most of the genAI stuff; explainer).

blog.rinesi.com

Marcelo Rinesi @marcelorinesi.bsky.social · 20h

As an aside, Argentinean journalists -and society- take most US coverage with the same skepticism they take AI industry coverage. The overall tone is "Chicago is dangerous" not "Trump is insane."

Marcelo Rinesi @marcelorinesi.bsky.social · 20h

Just saw an Argentinean sports program comment a breaking rumor that Monday's Argentina-Puerto Rico soccer friendly won't be played in [the] Chicago [War Zone] but in [Homeland] Miami.

#WorldCup rehearsals are going great 👍

1 1

Marcelo Rinesi @marcelorinesi.bsky.social · 21h

"The US is a democracy" and "the US is becoming less of a democracy week by week" are not contradictory but rather simultaneously true assessment.

Same for "Trump et al are insane clowns that can be defeated" and "Trump et al are not, at present, being defeated."

via @wikisteff.bsky.social

Rep. Dave Min @congressmin.bsky.social · 1d

It's been two weeks since Adelita Grijalva was elected and Mike Johnson has still refused to seat her, unlike in past cases where he swore in new members within 24 hours.

There cannot be one set of rules for Rs and another for Ds.

Also, release the damn Epstein files already.

Marcelo Rinesi @marcelorinesi.bsky.social · 1d

For one, the Trump admin feels at least as momentous as the fall of the Berlin Wall. But also the energy transition seems a sort of threshold: the (arguably) most important technological shift of our generation is not being led by the US but by China.

Of course, cf SImpsons' "... so far" meme.

1

Marcelo Rinesi @marcelorinesi.bsky.social · 1d

This mentions some clever ideas like zero-knowledge proofs but the politics and incentives don't add up. It kind of looks as if the IMF doesn't like stablecoins, knows the US is going to push them anyway and is trying to get in front of that parade and hopefully nudge it away from the worst ditches.

The Stablecoin Balancing Act

Fighting financial crime doesn’t have to come at the cost of privacy

www.imf.org

1 2

Marcelo Rinesi @marcelorinesi.bsky.social · 1d

Not gonna lie, 2025 is starting to look a bit 1989, geopolitical milestones-wise.

Renewables overtake coal as world's biggest source of electricity

Developing countries lead the historic clean energy charge but the US and EU rely more on fossil fuels than before, a think tank study shows.

www.bbc.com

1 1 5

Marcelo Rinesi @marcelorinesi.bsky.social · 1d

“The last act is the seldom forgiven: To get the right conclusion from the self-serving given.”

-- T. S. Eliot,
"Begging the Question in the Cathedral" 2/2

1

Marcelo Rinesi @marcelorinesi.bsky.social · 1d

I'm ambivalent. On one hand: yes, India has a grid problem (so does a *lot* of countries; in a cheap renewables world, it might be *the* problem). OTOH, this is a Goldman Sachs report talking about and assuming bubbleworths of AI demand, so... 1/

India's Grid Cannot Keep Up With Its Ambitions

The digital economy is arriving faster than the physical grid can adapt to handle its demands.

indiadispatch.com

1 1

Marcelo Rinesi @marcelorinesi.bsky.social · 1d

Legal restraints are academic if you can just make up grotesquely obvious lies and DOJ/FBI/ICE/the military will act on them, SCOTUS will have your back, and most of media will either amplify them or quote experts about pushing norms. 2/2

Marcelo Rinesi @marcelorinesi.bsky.social · 1d

They did try in J6. If you can pretend Portland is a war zone and send troops then the epistemic guardrails are gone: just declare state results invalid due to "buses of illegals" or whatever. No need to be subtle or sane. 1/

via @alongcamejones.bsky.social

G Elliott Morris @gelliottmorris.com · 1d

at this point you have to acknowledge that there is no way that these guys don't try to mess with the 2026 midterms. there is absolutely no respect for the legitimacy of the opposition

Blue Georgia @bluegeorgia.bsky.social · 1d

Stephen Miller: "Eventually elections don’t carry the day because violence beats votes… When we see a muscular response from the federal government, it’s because we’re not going to let violence beat votes."

Is he implying federal intervention determines political legitimacy, not ballots or courts?

1 1

Marcelo Rinesi @marcelorinesi.bsky.social · 1d

#RisingOnArxiv (ignoring a lot of genAI stuff):

* ICCV 2025
* Fermion sign problem
* Genuine multipartite entanglement

Rising on arXiv - 2025-10-05

A few of the ideas, topics, and commonplaces that have been gaining steam on arXiv during the last few months (yes, I do a lot of filtering to keep away most of the genAI stuff; explainer).

blog.rinesi.com

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

#RisingOnArxiv (ignoring a lot of genAI stuff):

* ICCV 2025
* Fermion sign problem
* Genuine multipartite entanglement

Rising on arXiv - 2025-10-05

blog.rinesi.com

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

... the technological and cultural difficulties to be managed, but what I'm really looking forward to is the move from IMO-style problems to the sort of work done, e.g., in mathematical physics or abstract mathematics. Speed up *that* work, and you change the speed at which the world changes. 2/2

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

By @tachim.bsky.social et a long al. As a caveat, this is a report on a company product and it's an industry with an uneven record on such but I'm very partial to this architecture (generate drafts, formal engines for verification) so I'm happy to see this sort of development.

I understand... 1/

Aristotle: IMO-level Automated Theorem Proving

We introduce Aristotle, an AI system that combines formal verification with informal reasoning, achieving gold-medal-equivalent performance on the 2025 International Mathematical Olympiad problems. Ar...

arxiv.org

1

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

I like this one on first principles --one of my core bets is that biology is too complex to meaningfully think about with our existing set of collective intellectual tools-- but note that what makes this all possible is the existence of interventional data. You got to put in the work.

Large-Scale Bayesian Causal Discovery with Interventional Data

Inferring the causal relationships among a set of variables in the form of a directed acyclic graph (DAG) is an important but notoriously challenging problem. Recently, advancements in high-throughput...

arxiv.org

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

"On two occasions I have been asked, 'Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?' I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question." Many decision-makers are not very far from that position...5/5

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

... what you're observing isn't the world but the _implications_ of your [existing] model, and the network you're training is a tool to make some uses of that easier.

I'm being a bit pedantic, I know. But OTOH, there's that Babbage quote: 4/

1

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

... (your mechanism) in some other way and conceptually are building a more usable, if more opaque, surrogate through differentiable programming. "Synthetic data" is the usual term but IMHO it's misleading: 3/

1

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

There's often confusion on this because the prototypical use case of neural networks is to build models from observational data, but here the information flow is completely different: you already built your model... 2/

1

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

By Carson Dudley and @marisa-eisenberg.bsky.social

Besides the technical results this paper is important as a clarification of an often overlooked use case of neural networks: to facilitate working out the implications of a mechanism by building a surrogate model. 1/

Learning From Simulators: A Theory of Simulation-Grounded Learning

Simulation-Grounded Neural Networks (SGNNs) are predictive models trained entirely on synthetic data from mechanistic simulations. They have achieved state-of-the-art performance in domains where real...

arxiv.org

1

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

Mostly a null result, which is impressive; few groups are willing to say "so far meh" so much kudos to them[1] .

By @marthagimbel.bsky.social et al.

[1] Although I'm biased because this fits my own anecdotal observations and analysis.

Evaluating the Impact of AI on the Labor Market: Current State of Affairs

budgetlab.yale.edu

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

As a #Stan user I'm not thrilled by this, but Bob Carpenter says "Like Stan, JAX is also a differentiable programming language. Unlike Stan, it’s wonderfully compositional and general." and he knows a bit of a lot more about this than I do, so to the pipeline of things to look into JAX goes...

Andrew Gelman et al. @statmodeling.bsky.social · 4d

It’s a JAX, JAX, JAX, JAX World
statmodeling.stat.columbia.edu/2025/10/03/i...

It’s a JAX, JAX, JAX, JAX World | Statistical Modeling, Causal Inference, and Social Science

statmodeling.stat.columbia.edu

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

It's not the whole of it, but a political/media system built over decades to bothsides whether all climatologists are engaged in a vast conspiracy can as a side effect be rendered unable to say -or see- that the President is delusional.

via @kevinmkruse.bsky.social @estebanjq.bsky.social

Acyn @acyn.bsky.social · 3d

Trump: Portland is burning to the ground—insurrectionists all over the place. The politicians are afraid for their lives. That’s the only reason they say there’s nothing happening.

1

Marcelo Rinesi @marcelorinesi.bsky.social · 2d

Not saying "things will get bad if..." -things are already on fire! nothing is fine!- but once a soldier fires into a group of civilians and gets a pardon (even odds on a medal and/or whoever punishes them being reprimanded) you unlock a new and deeper circle of bad.

via @robshum.bsky.social

sixmile.bsky.social @sixmile.bsky.social · 3d

As a dear friend reminded me: Stephen "Goebbels" Miller recently said that law enforcement should act lethally and not worry about it because Trump will pardon them. We're seeing behavior shielded by impunity that belies any suggestion they will ever leave power because of a lost election.

2