Marciano Siniscalchi @marciano.bsky.social · 04/10/2026Nobody does school protests like France. 010 1
Marciano Siniscalchi @marciano.bsky.social · 16/09/2026Thanks for giving me the opportunity to re-read that gem of a take-down. 000 0
Reposted by Marciano SiniscalchiPhillips OBrien @phillipspobrien.bsky.social · 16/09/2026Latest in @theatlantic.com . What we are seeing now is US weakness across the board. The American Age is over. What comes next will be very different. Gift link. www.theatlantic.com/ideas/2026/0...theatlantic.comThe End of the American AgeThe Iran war marks the conclusion of eight decades of global dominance. 22382166 382
Reposted by Marciano SiniscalchiKieran Healy @kjhealy.co · 16/09/2026They should have taken the opportunity to combine the Stanford Prison Experiment, the Stanford Marshmallow Experiment, and the Stanford Linear Accelerator into a show where people assigned as guards or prisoners are repeatedly bombarded with either one marshmallow now or two marshmallows later. 717427 174
Marciano Siniscalchi @marciano.bsky.social · 13/09/2026Eating soft pretzels when flying to Italy via FRA or MUC is a beloved ritual of the Siniscalchi family 020 2
Marciano Siniscalchi @marciano.bsky.social · 12/09/2026yoshuabengio.org/en/publicati...yoshuabengio.orgYoshua Bengio | Why are AI agents lying, cheating and coordinating?A lot has been written about the incidents of the last few months in which AI agents misbehaved in serious ways. They took actions that would be considered as crimes if a human took them, escaped thei... 011 1
Marciano Siniscalchi @marciano.bsky.social · 30/08/2026Honestly, no. He's savvy enough to know that such a comment would be "misconstrued." Hence, he meant what he said, and that's not reassuring. 051 5
Marciano Siniscalchi @marciano.bsky.social · 27/08/2026Robot arms, what could possibly go wrong 020 2
Marciano Siniscalchi @marciano.bsky.social · 26/08/2026Honestly, I find this sort of PR (because that's what it is) really puzzling. OpenAI is basically telling its potential customers that they can't be trusted to secure their own systems. How can said customers feel comfortable sending their most sensitive data to OpenAI's models? 010 1
Reposted by Marciano SiniscalchiRetraction Watch @retractionwatch.com · 24/08/2026Six editors and nearly 60 advisory editors of Games and Economic Behavior resigned after Elsevier, the journal’s publisher, decided not to renew the editor-in-chief’s contract.retractionwatch.comEditors of economics journal resign after Elsevier doesn’t renew EIC’s contractSix editors and nearly 60 advisory editors of Games and Economic Behavior resigned after the journal’s publisher decided not to renew the editor-in-chief’s contract. In late July, Elsevier to… 03818 38
Marciano Siniscalchi @marciano.bsky.social · 21/08/2026I never knew bond trading could be such a dangerous profession 020 2
Marciano Siniscalchi @marciano.bsky.social · 21/08/2026What the post should say: "...a NORTHWESTERN UNIVERSITY expert and others say" (the guy in the picture) 140 4
Marciano Siniscalchi @marciano.bsky.social · 20/08/2026Agree. That said, I found the article a bit shallow. Worse, it didn't even acknowledge the existence of Daron memes 🙂 030 3
Marciano Siniscalchi @marciano.bsky.social · 15/08/2026Why R? Just hand-code your matrix algebra in ARM assembly 🤣 130 3
Marciano Siniscalchi @marciano.bsky.social · 15/08/2026Actually, come to think about it, you're quite right. Many of us literally grew up in Milan, too! I can certainly think of Bocconi people who are not decision theorists and did not grow up in Milan, but as far as (twentieth-century) decision theorists are concerned... hmm, interesting! 120 2
Marciano Siniscalchi @marciano.bsky.social · 15/08/2026Also, "from Milan" is a bit ambiguous. "From Bocconi" would be more accurate (and correct for us GenXers). Yes, contrary to popular belief, there are other universities (with econ departments) in Milan! 120 2
Marciano Siniscalchi @marciano.bsky.social · 14/08/2026I'm not really aware of anything worth reporting... things may be different now, but growing up in Milan, there were relatively few Chinese restaurants in Milan, and none of them good, so I went very rarely. I remember very standard fare, nothing imaginative (even in a bad way hahaha!) 210 2
Marciano Siniscalchi @marciano.bsky.social · 13/08/2026I say we repatriate all Italian gelatai from Germany, pronto. That'll teach them! 220 2
Marciano Siniscalchi @marciano.bsky.social · 13/08/2026I read it... I also signed up... still a bit confused, sorry :-( 100 1
Marciano Siniscalchi @marciano.bsky.social · 13/08/2026Please ELI5: how exactly is this different from following a science-related feed (other than LaTeX - yay! - and longer posts)? That is, do I get to follow different people on Lea and on Bsky? Do my posts on Lea appear on Bsky too and vice versa? Maybe a short explainer might help others too! Thanks! 110 1
Marciano Siniscalchi @marciano.bsky.social · 11/08/2026@lopatto.bsky.social read Zuck's 6,500-word screed so we don't have to 020 2
Marciano Siniscalchi @marciano.bsky.social · 08/08/2026I agree 100% with this. We are still learning how to best use these tools, and so it makes sense to take a cautious, yet open-minded approach. 040 4
Marciano Siniscalchi @marciano.bsky.social · 07/08/2026Also, why are these "revelations" not accompanied by patently incompetent security "experts" being fired? 010 1
Marciano Siniscalchi @marciano.bsky.social · 07/08/2026These are all fair points, but just to clear the air, the idea AFAIK is simply to provide the AI version of "technical referees" whose task is to spot issues with proofs, etc. What difference does it make that it's an AI rather than a human (other than, potentially, the quality of the review)? 010 1
Marciano Siniscalchi @marciano.bsky.social · 07/08/2026Are you worried that referees will be less careful about technical details? It's possible, but the fact that these tools are used after conditional acceptance makes things ... interesting: what is the AI catches an obvious flaw that you, a referee known to the editor, did not spot? 100 1
Marciano Siniscalchi @marciano.bsky.social · 06/08/2026Also, Refine reports will be treated similarly to regular referee reports: editors can highlight only certain points, and of course authors can object to the points raised. BTW, since AI is intrinsically random, running Refine twice need not catch the same issues (except for glaring ones). 2/2 120 2
Marciano Siniscalchi @marciano.bsky.social · 06/08/2026To clarify, as far as ECMA is concerned, Refine reports will be offered at the cond'l acceptance stage, so there is no strategic advantage to running Refine before submission (though I would personally use it-or similar tools-to improve quality). 1/2 220 2
Marciano Siniscalchi @marciano.bsky.social · 05/08/2026In principle, sure, but the economics are against it. Suppose you use an LLM to go from prompt to binary, and spend X dollars in tokens. If you literally only keep the prompt and the binary (no intermediate product) every time you want to add a feature, your cost is again on the order of X dollars. 020 2
Marciano Siniscalchi @marciano.bsky.social · 26/07/2026For a moment I thought I had missed some deep literary or philosophical reference 030 3
Marciano Siniscalchi @marciano.bsky.social · 24/07/2026...but maybe there is a way: www.reddit.com/r/Amazing/s/...reddit.comFrom the Amazing community on Reddit: 32/35 students failed this professor's midterm using AI because he secretly hid some white font in his text.Explore this post and more from the Amazing community 010 1
Marciano Siniscalchi @marciano.bsky.social · 24/07/2026It's even worse. I was talking about this with a colleague in Italy. Even with an in-class exam, you need to worry about hidden second phones (yes), watches, even glasses. Her institution considered blocking cell signals but apparently that's illegal bc it can cause interference outside the building 110 1
Reposted by Marciano SiniscalchiJames Bland @jamesbland.bsky.social · 21/07/2026🚨🚨🚨 New working paper! 🚨🚨🚨 "The Hunt for Loss Aversion: A Bayesian Detective Story" with Brian Monroe and Glenn Harrison #EconSky #ProspectTheory #Bayes cear.gsu.edu/wp-2026-01-t... a thread ... 291 9
Reposted by Marciano Siniscalchiandrew.ottawa 🇨🇦 @ottawandrew.bsky.social · 19/07/2026THE HALFTIME SHOW WILL CONTINUE UNTIL MORALE IMPROVES. #ESPARG #WorldCup 142 4
Reposted by Marciano SiniscalchiDevon Heinen @devonheinen.bsky.social · 19/07/2026#Bieber was the hydration break of the #WorldCup #HalftimeShow 011 1