- nexa - server-nexa.polito.it

When billion-dollar AIs break down over puzzles a child can do, it’s time to rethink the hype
by Daniela Tafani June 13, 2025

June 13, 2025

When billion-dollar AIs break down over puzzles a child can do, it’s time to rethink the hype Gary Marcus The tech world is reeling from a paper that shows the powers of a new generation of AI have been wildly oversold A research paper by Apple has taken the tech world by storm, all but eviscerating the popular notion that large language models (LLMs, and their newest variant, LRMs, large reasoning models) are able to reason reliably. Some are shocked by it, some are not. The well-known venture capitalist Josh Wolfe went so far as to post on X that “Apple [had] just GaryMarcus’d LLM reasoning ability” – coining a new verb (and a compliment to me), referring to “the act of critically exposing or debunking the overhyped capabilities of artificial intelligence … by highlighting their limitations in reasoning, understanding, or general intelligence”. Apple did this by showing that leading models such as ChatGPT, Claude and Deepseek may “look smart – but when complexity rises, they collapse”. In short, these models are very good at a kind of pattern recognition, but often fail when they encounter novelty that forces them beyond the limits of their training, despite being, as the paper notes, “explicitly designed for reasoning tasks”. As discussed later, there is a loose end that the paper doesn’t tie up, but on the whole, its force is undeniable. So much so that LLM advocates are already partly conceding the blow while hinting at, or at least hoping for, happier futures ahead. In many ways the paper echoes and amplifies an argument that I have been making since 1998: neural networks of various kinds can generalise within a distribution of data they are exposed to, but their generalisations tend to break down beyond that distribution. A simple example of this is that I once trained an older model to solve a very basic mathematical equation using only even-numbered training data. The model was able to generalise a little bit: solve for even numbers it hadn’t seen before, but unable to do so for problems where the answer was an odd number. More than a quarter of a century later, when a task is close to the training data, these systems work pretty well. But as they stray further away from that data, they often break down, as they did in the Apple paper’s more stringent tests. Such limits arguably remain the single most important serious weakness in LLMs. The hope, as always, has been that “scaling” the models by making them bigger, would solve these problems. The new Apple paper resoundingly rebuts these hopes. They challenged some of the latest, greatest, most expensive models with classic puzzles, such as the Tower of Hanoi – and found that deep problems lingered. Combined with numerous hugely expensive failures in efforts to build GPT-5 level systems, this is very bad news. The Tower of Hanoi is a classic game with three pegs and multiple discs, in which you need to move all the discs on the left peg to the right peg, never stacking a larger disc on top of a smaller one. With practice, though, a bright (and patient) seven-year-old can do it. What Apple found was that leading generative models could barely do seven discs, getting less than 80% accuracy, and pretty much can’t get scenarios with eight discs correct at all. It is truly embarrassing that LLMs cannot reliably solve Hanoi. And, as the paper’s co-lead-author Iman Mirzadeh told me via DM, “it’s not just about ‘solving’ the puzzle. We have an experiment where we give the solution algorithm to the model, and [the model still failed] … based on what we observe from their thoughts, their process is not logical and intelligent”. The new paper also echoes and amplifies several arguments that Arizona State University computer scientist Subbarao Kambhampati has been making about the newly popular LRMs. He has observed that people tend to anthropomorphise these systems, to assume they use something resembling “steps a human might take when solving a challenging problem”. And he has previously shown that in fact they have the same kind of problem that Apple documents. If you can’t use a billion-dollar AI system to solve a problem that Herb Simon (one of the actual godfathers of AI) solved with classical (but out of fashion) AI techniques in 1957, the chances that models such as Claude or o3 are going to reach artificial general intelligence (AGI) seem truly remote. So what’s the loose thread that I warn you about? Well, humans aren’t perfect either. On a puzzle like Hanoi, ordinary humans actually have a bunch of (well-known) limits that somewhat parallel what the Apple team discovered. Many (not all) humans screw up on versions of the Tower of Hanoi with eight discs. But look, that’s why we invented computers, and for that matter calculators: to reliably compute solutions to large, tedious problems. AGI shouldn’t be about perfectly replicating a human, it should be about combining the best of both worlds; human adaptiveness with computational brute force and reliability. We don’t want an AGI that fails to “carry the one” in basic arithmetic just because sometimes humans do. Whenever people ask me why I actually like AI (contrary to the widespread myth that I am against it), and think that future forms of AI (though not necessarily generative AI systems such as LLMs) may ultimately be of great benefit to humanity, I point to the advances in science and technology we might make if we could combine the causal reasoning abilities of our best scientists with the sheer compute power of modern digital computers. What the Apple paper shows, most fundamentally, regardless of how you define AGI, is that these LLMs that have generated so much hype are no substitute for good, well-specified conventional algorithms. (They also can’t play chess as well as conventional algorithms, can’t fold proteins like special-purpose neurosymbolic hybrids, can’t run databases as well as conventional databases, etc.) What this means for business is that you can’t simply drop o3 or Claude into some complex problem and expect them to work reliably. What it means for society is that we can never fully trust generative AI; its outputs are just too hit-or-miss. One of the most striking findings in the new paper was that an LLM may well work in an easy test set (such as Hanoi with four discs) and seduce you into thinking it has built a proper, generalisable solution when it has not. To be sure, LLMs will continue to have their uses, especially for coding and brainstorming and writing, with humans in the loop. But anybody who thinks LLMs are a direct route to the sort of AGI that could fundamentally transform society for the good is kidding themselves. <https://www.theguardian.com/commentisfree/2025/jun/10/billion-dollar-ai-puz…>

4 4

L'ai che mangia se stessa
by don Luca Peyron June 13, 2025

June 13, 2025

https://www.theguardian.com/music/2025/jun/03/ai-bot-farms-and-innocent-ind… -- _________________________ don Luca Peyron Pastorale Universitaria - Apostolato Digitale Arcidiocesi di Torino www.universitari.to.it via XX settembre 83, Torino tel. 011 5156239

1 0

Impact of Military Artificial Intelligence on Nuclear Escalation Risk - SIPRI Report
by Diego Latella June 13, 2025

June 13, 2025

FYI Impact of Military Artificial Intelligence on Nuclear Escalation Risk Vladislav Chernavskikh and Jules Palayer June 2025 Stockholm SIPRI https://doi.org/10.55163/FZIW8544 DL =================== Dott. Diego Latella - Formerly Senior Researcher with CNR/ISTI (retired)

1 0

RIPROGRAMMATO il 184° Mercoledì di Nexa | 25 giugno 2025, ore 17.00
by Nexa - Media June 12, 2025

June 12, 2025

Gentilissime, gentilissimi, Vi segnaliamo che il 184° Mercoledì di Nexa "A Law and Political Economy of Intellectual Property", con ospite Talha Syed (UC Berkeley School of Law), è stato riprogrammato in data 25 giugno 2025, dalle ore 17.00 alle ore 19.00. L'incontro si terrà IN PRESENZA (presso la sede del Centro Nexa) e ONLINE (https://didattica.polito.it/VClass/NexaEvent) Maggiori informazioni sono disponibili alla pagina: https://nexa.polito.it/mercoledi-184 Un saluto cordiale, -- Valeria Bergantino Communication Manager Nexa Center for Internet & Society Politecnico di Torino - DAUIN Via Pier Carlo Boggio, 65/A - 10138 Torino web: https://nexa.polito.it/ mail: valeria.bergantino(a)polito.it<mailto:valeria.bergantino@polito.it> tel: 3473443585

1 0

ChatControl e "Going Dark"
by J.C. DE MARTIN June 11, 2025

June 11, 2025

Certi interessi hanno una perseveranza davvero ammirevole... passano i lustri, anzi, i decenni, ma loro, talpone pazienti e tenaci, continuano a scavare, a toglierci la terra da sotto i piedi... vediamo di fare un po' di rumore, di mobilitare qualche parlamentare europeo... juan carlos - *EU Going Dark: il piano di SORVEGLIANZA DI MASSA dell'Unione Europea*. Breve video (10') di Morrolinux: pacato, lucido, incisivo:https://youtu.be/cMxwjegG7ys?si=wXZSB7nArKN6ikEU *- Impact assessment on retention of data by service providers for criminal proceedings* https://ec.europa.eu/info/law/better-regulation/have-your-say/initiatives/1… <https://ec.europa.eu/info/law/better-regulation/have-your-say/initiatives/1…> Scadenza per dare feedback: *_18 giugno 2025_* (mezzanotte ora di Bruxelles). - *High-Level Group (HLG) on access to data for effective law enforcement* a.k.a. *HLG "Going Dark"* (non scherzo, il loro indirizzo email ufficiale è: EC-HLG-GOING-DARK(a)ec.europa.eu) https://home-affairs.ec.europa.eu/networks/high-level-group-hlg-access-data… <https://home-affairs.ec.europa.eu/networks/high-level-group-hlg-access-data…> /HLG Recommendations/: https://home-affairs.ec.europa.eu/document/download/1105a0ef-535c-44a7-a6d4… /Concluding report/: https://home-affairs.ec.europa.eu/document/download/4802e306-c364-4154-835b… - EDRI, High-Level Group “Going Dark” outcome: A mission failure <https://edri.org/our-work/high-level-group-going-dark-outcome-a-mission-fai…> (13 June 2024) - EDRI, ‘ProtectEU’ security strategy: a step further towards a digital dystopian future <https://edri.org/our-work/protecteu-security-strategy-a-step-further-toward…> (30 April 2025) - MEP Patrick Breyer, Going Dark expert group – EU’s surveillance forge <https://www.patrick-breyer.de/en/posts/going-dark-expert-group-eus-surveill…> - Per firmare contro Chat Control: https://stopchatcontrol.eu/ jc

2 1

ANNULLATO il 184° Mercoledì di Nexa
by Nexa - Media June 10, 2025

June 10, 2025

Gentilissime, gentilissimi, Vi segnaliamo che il 184° Mercoledì di Nexa dal titolo "A Law and Political Economy of Intellectual Property", in programma per domani, è stato annullato. L'incontro verrà riprogrammato in data da destinarsi a breve. Un saluto cordiale, -- Valeria Bergantino Communication Manager Nexa Center for Internet & Society Politecnico di Torino - DAUIN Via Pier Carlo Boggio, 65/A - 10138 Torino web: https://nexa.polito.it/ mail: valeria.bergantino(a)polito.it<mailto:valeria.bergantino@polito.it> tel: 3473443585

1 0

Pubblicazione report relativo al progetto "Artificial Intelligence between ‘Openness’ and Open Source"
by Nexa - Media June 10, 2025

June 10, 2025

Gentilissime, gentilissimi, È con piacere che vi segnaliamo che è stato recentemente pubblicato un nuovo report relativo al progetto dal titolo “Artificial Intelligence between ‘Openness’ and Open Source”. Lo studio, frutto di una collaborazione tra il Centro Nexa su Internet & Società e TIM S.p.A., analizza le evoluzioni e le problematiche legate all'applicazione del modello open source nel contesto dell'intelligenza artificiale, con particolare riferimento al settore delle telecomunicazioni, soffermandosi su recenti iniziative legislative come l'AI Act per valutarne efficacia ed implicazioni. Il testo completo del report, insieme a maggiori informazioni in merito al progetto, è disponibile al seguente link: https://nexa.polito.it/artificial-intelligence-between-openness-and-open-so… Vi ringraziamo per l’attenzione. Cordiali saluti, -- Valeria Bergantino Communication Manager Nexa Center for Internet & Society Politecnico di Torino - DAUIN Via Pier Carlo Boggio, 65/A - 10138 Torino web: https://nexa.polito.it/ mail: valeria.bergantino(a)polito.it<mailto:valeria.bergantino@polito.it> tel: 3473443585

1 0

The Problem With "AI Fluency"
by Daniela Tafani June 9, 2025

June 9, 2025

The Problem With "AI Fluency" There is a growing tech trend in which questioning AI and recognizing that it's not the right tool for every task are no longer welcome Luiza Jarovsky Jun 09, 2025 As the “AI-first” narrative gains more traction in the tech industry, a few days ago, a post from the CEO of Zapier describing how the company measures “AI fluency” went viral. In today's edition, I discuss Zapier's approach and the problem with the expectation of “AI fluency” that is spreading within tech companies and beyond, as it might be harmful and could backfire, both legally and ethically. - In addition to the memos, leaked emails, and PR announcements showing how each company is prioritizing AI (Meta, Duolingo, Shopify, and Fiverr are recent examples - read my recent analysis), there are also worldwide “AI fluency” efforts aiming to incentivize employees to use AI no matter what. If you search “AI fluency” on LinkedIn, you'll see that there is a growing number of posts about the topic, as well as professionals who have added this skill or title to their profiles. There are also job openings that mention or require it, showing that it's a terminology that is becoming normalized and an integral part of the current AI zeitgeist. Before I continue, a reminder that learning about AI, upskilling, and integrating it professionally is extremely important. I wrote about AI literacy multiple times in this newsletter, including what laws, such as the EU AI Act, say about it. AI literacy, however, demands critical thinking, as well as ethical and legal awareness, including the ability to know when not to use AI. This is the opposite of what recent corporate approaches to “AI fluency” are promoting. Let's take a look at Zapier. A few days ago, their CEO posted the table below. It shows how they assess AI fluency by role and maps skills across four levels (unacceptable, capable, adoptive, and transformative): In the post, he also shared AI-related questions Zapier has been asking during interviews, and highlighted that requirements are role-specific and “the bar will keep rising.” Let me start by saying that when discussing “fluency,” the four categories necessarily involve a value judgment: "unacceptable" is worse than "capable", which is worse than "adoptive", which is worse than "transformative". The optimal column is the last one (“transformative”). With that in mind, pay attention to the "unacceptable" column. Across different roles and in various contexts, the employee described in this column is someone who is generally skeptical of AI tools and avoids them. This is an employee who will likely be fired soon, or a job applicant who will not be hired as they do not fit Zapier's expectations. Pay attention to how this doesn't necessarily mean the person doesn't understand AI, or doesn't know how to use it. In fact, this person might be consciously rejecting AI (or being critical of it) for a specific task because they know AI: will lead to inaccurate outputs will lead to low-quality outcomes is proven not to be the best tool to accomplish that specific task is legally risky for that specific task might require additional safety or ethical checks [whatever other perfectly legitimate reason to reject AI deployment in a specific professional context] Zapier and other companies spreading the AI fluency dogma don't want to hear about that and don't accept that AI will be rejected, even if it's for a legitimate reason. Critical thinking around AI, and the understanding that it's an automating tool that is not suitable for all tasks, is not welcome, even though we have more and more studies showing its limitations (the latest one from Apple). For Zapier and other companies promoting similar AI fluency approaches, AI is always good, and it must be accepted, implemented, and praised. Zapier is a private company, and its management team is free to act as it wishes within legal boundaries. This type of corporate culture, however, can be harmful. Why? <https://www.luizasnewsletter.com/p/the-problem-with-ai-fluency>

1 0

MEMENTO | 184° Mercoledì di Nexa | 11 giugno 2025, ore 17.00
by Nexa - Media June 9, 2025

June 9, 2025

Gentilissime, gentilissimi, Vi ricordiamo che mercoledì 11 giugno, alle ore 17.00, si terrà il 184° Mercoledì di Nexa con un incontro dal titolo "A Law and Political Economy of Intellectual Property". Ospite dell'incontro: Talha Syed (UC Berkeley School of Law). L'incontro si terrà IN PRESENZA e ONLINE, in lingua inglese. SEDE FISICA dell'incontro: Centro Nexa su Internet e Società, Politecnico di Torino, Via Boggio 65/a, Torino (1° piano). Per accedere alla sala si raccomanda di suonare al citofono Portineria e di seguire le indicazioni segnalate lungo il percorso. QUI<https://nexa.polito.it/contatti> maggiori informazioni su come raggiungerci. STANZA VIRTUALE dell'incontro: https://didattica.polito.it/VClass/NexaEvent Maggiori informazioni alla pagina: https://nexa.polito.it/mercoledi-184 Un saluto cordiale, -- Valeria Bergantino Communication Manager Nexa Center for Internet & Society Politecnico di Torino - DAUIN Via Pier Carlo Boggio, 65/A - 10138 Torino web: https://nexa.polito.it/ mail: valeria.bergantino(a)polito.it<mailto:valeria.bergantino@polito.it> tel: 3473443585

1 0

What Trump’s and Musk’s Public Divorce Tells Us About the State of Tech
by M. Fioretti June 7, 2025

June 7, 2025

Il mio commento, per FOSS Force https://fossforce.com/2025/06/what-trumps-and-musks-public-divorce-tells-us… in sintesi: questa farsa e' ESATTAMENTE come il divorzio fra Trump e Ivana del 1992, ma ricorda e prova due cose, mai ripetute abbastanza: 1) in democrazie serie, i politici non dovrebbero MAI parlare attraverso i social media 2) quanto le tecnologie chiuse mettono tutti alla merce' di pochissime persone, o addirittura una, in questo caso Marco

1 0