- nexa - server-nexa.polito.it

La pagina bianca è la mia bottega (di Michele Serra)
by Antonio June 20, 2025

June 20, 2025

"Scrivi una favola della buonanotte sui dinosauri per un bambino di otto anni". "Progetta un piano turistico di due settimane nel parco di Yellowstone". "Scrivi una guida di un colloquio per una posizione interna di ingegnere informatico". Queste tre scritte, assieme ad altre dalla logica misteriosa, compaiono da qualche settimana, dentro tre finestrelle rettangolari, in testa alla pagina Word che ogni giorno apro nel mio Mac per mettermi a scrivere. La pagina bianca è la mia bottega. Il luogo, privatissimo, del mio lavoro quotidiano. Da qualche mese la AI di Microsoft la presidia senza scampo. Apro la porta della mia bottega e trovo un intruso che mi dice "buongiorno, posso esserti utile?". È entrato da solo, senza che io l'abbia mai chiesto, proponendomi di fare in mia vece ciò che faccio da una vita intera. Ammesso (e non concesso) che se le chiedessi di scrivere questa newsletter al mio posto - stamattina fa troppo caldo, non ho voglia di faticare e preferisco andare al fiume - la scriverebbe meglio di me, il problema è che non gliel'ho chiesto; né voglio chiederlo. Non voglio quei banner in cima alla mia pagina: io, nella mia pagina, voglio stare da solo. Oppure sarebbe normale, secondo voi, che il disegnatore aprisse la risma appena comperata in cartoleria e trovasse in ogni foglio un segno già tracciato a matita? Con la scritta: se vuoi puoi cancellarmi, ma non ti sembro un buon inizio per quello che intendi disegnare? Conosco l'obiezione. Basta cominciare a scrivere - basta un clic - e quegli inviti non richiesti a servirsi di un coautore spariscono (e ci mancherebbe altro). Il foglio torna intonso. Ma obietto a mia volta che se io volessi tornare ad aprire la mia pagina trovandola, come sempre, immacolata, libera da ogni intrusione, non ne sarei capace. Importante: non vale alcun livello di biasimo per il mio "non essere capace". Niente e nessuno può obbligarmi (e obbligare chiunque) all'incessante aggiornamento sulle novità tecnologiche; è una mia facoltà, non un mio dovere. Uso quanto, della tecnologia, mi è utile e necessario, fosse anche un livello "primitivo" rispetto al ricco catalogo disponibile. Perché dovrei sentirmi costretto ad attrezzarmi per continui salti di qualità? Volessi inibire le intrusioni indebite nel mio programma di scrittura - che già così com'è soddisfa perfettamente le mie necessità - dovrei rivolgermi a un professionista o a un amico nerd che lo faccia al mi o posto: come se il disegnatore di cui sopra non avesse la gomma per cancellare, e dovesse chiedere a qualcun altro di pulirgli il foglio. Ogni volta che parlo della questione con gli amici, si accende il dibattito. I punti forti del pensiero di chi ritiene eccessivo il mio disappunto sono questi: 1) È sempre accaduto che i salti tecnologici producessero sconcerto e spiazzassero i meno giovani: poi ci si assesta, si capisce che i vantaggi sono superiori agli svantaggi, e ciò che sembrava sconquassante diventa del tutto normale. 2) Non si può pretendere che la propria arretratezza tecnologica pesi al punto da attardare il progresso. Se io sono un troglodita digitale non ho il diritto di prendermela con chi, non essendolo, maneggia le nuove situazioni con la giusta destrezza. Ben detto. Convincente. Ma non è questo il punto. Il punto è che l’adesione al progresso – dando per scontato che l’IA lo sia – non può essere obbligatoria. Chi andava a piedi poi ha cominciato a usare la bicicletta e chi andava in bicicletta poi ha cominciato a usare l’automobile. E sicuramente già quei passaggi d’epoca furono accompagnati dal borbottio passatista, “dove andremo a finire, quei velocipedi sono pericolosissimi!”. Ma si è continuato ad andare a piedi, e in bicicletta, anche nell’epoca dell’automobile (e anzi, con il tempo, camminare e pedalare hanno riconquistato un’aura formidabile di salubrità e, aggiungo, di “modernità”). Le nuove tecnologie si sono sempre aggiunte, mai sostituite alle precedenti. Nessun pedone si è ritrovato una bicicletta in casa senza averla richiesta, nessun ciclista si è ritrovato un’automobile davanti alla porta senza preavviso. Bene: io non sono affatto sicuro di poter continuare a lavorare su una pagina bianca, come vorrei e come avrei diritto di fare. La pervasività e l’intrusività delle tecnologie digitali mi fa dubitare, ogni giorno di più, che siano facoltative: e già adesso, vederle comparire sul mio schermo senza che io abbia fatto nulla per convocarle, non mi piace. E anzi: mi turba. La vivo come una violenza nemmeno tanto subdola: e anzi evidente. Non perché io sia troglodita (lo sono, ma non più di tanto; non al punto di desiderare una clava per bastonare i chip), ma perché non voglio essere il terminale inerte di decisioni e di politiche commerciali che considerano “naturale” la mia adesione, e scontato il mio entusiasmo. Non voglio sentirmi reclutato a mia insaputa, e considero mio diritto, nel caso non voglia adeguarmi (o non sia in grado di farlo), essere lasciato in pace. Aggiungo: anche se gli accessi al “nuovo” fossero del tutto gratuiti – e non lo sono, voglio avere il diritto di non considerarli indispensabili alla mia vita e alle mie necessità. Il giorno che avrò bisogno di un'intelligenza artificiale, cosiddetta, voglio essere io a dirglielo. La rete che ci avvolge dovrebbe essere al nostro servizio, non noi al suo. Non nasciamo clienti, lo diventiamo secondo volontà e bisogni che è bene siano stabiliti da noi e non imposti dall’alto (dall’alto? Con il digitale viene voglia di dire: da ovunque, da sopra, da sotto, da “dentro”…). Vogliamo andare a piedi anche se abbiamo l’automobile e mentre camminiamo non vogliamo essere affiancati da un’automobile a guida automatica che ci dice “e dai, sali, perché ti ostini a fare tutta quella fatica?”. Grazie, si levi di torno, non ha idea di quanto mi piaccia camminare. E non ha idea di quanto sia sgradevole essere disturbato mentre cammino. “La pervasività e l’intrusività delle tecnologie digitali mi fa dubitare, ogni giorno di più, che siano facoltative: e già adesso, vederle comparire sul mio schermo senza che io abbia fatto nulla per convocarle, non mi piace”

9 16

Disdire iscrizione a Nexa
by Giulio Priano June 19, 2025

June 19, 2025

Buonasera, Scrivo la presente per richiedere la mia disiscrizione a Nexa.

2 1

[GI-STS] WEBINAR REMINDER - Climate and Humanitarian Consequences of an even Limited Nuclear Exchange and the Actual Risks of Nuclear War
by diego latella June 18, 2025

June 18, 2025

*** PLEASE DISSEMINATE *** "Climate and Humanitarian Consequences of an even Limited Nuclear Exchange and the Actual Risks of Nuclear War" Open webinar – June 26 – 4pm (CET) with David Ellwood (Council of the Pugwash Conferences on Science and World Affairs) "Climate and Humanitarian Consequences of an even Limited Nuclear Exchange" Paolo Cotta Ramusino (Former Secretary General of Pugwash Conferences on Science and World Affairs) "The Actual Risks of Nuclear War" Moderated by Mieke Massink - CNR ISTI; GI-STS, Pisa (The official language of the webinar is English) The event is organized by: Gruppo Interdisciplinare su Scienza, Tecnologia e Società (GI-STS) dell’Area della Ricerca di Pisa del CNR In cooperation with: Areaperta – Area della Ricerca CNR di Pisa Caffè della Scienza “N. Badaloni” - Livorno Centro Interdisciplinare Scienze per La Pace dell’Università di Pisa Istituto di Biofisica del CNR Istituto di Fisiologia Clinica del CNR Istituto di Scienza e Tecnologie dell’Informazione ``A. Faedo’’ del CNR La Nuova Limonaia Laboratorio Informatica e Società del CINI Pugwash Conferences on Science and World Affairs Unione degli Scienziati Per Il Disarmo For more information (including the link for participation, please look at: http://gists.pi.cnr.it/Events/2025_06_26_Pisa/2025_06_26_Pisa_main.htm D. Latella -- Dott. Diego Latella - Formerly Senior Researcher with CNR/ISTI - Now retired

1 0

ridefinire l'umano
by alessandro marzocchi June 17, 2025

June 17, 2025

Ridefinire l'umano, paradossalmente credo sia questo il passo avanti della macchina, oggi l'IA. Anche a proposito della "intelligenza", IA e ricerche su comunicazione e "mente" degli animali [1] ci aiutano ad individuare e capire componenti che prima erano indistinte, indefinite. Fermo restando che la macchina non ha corpo, con tutto quanto consegue, aumentano le attività che prima erano considerate esclusivamente umane e che diventano eseguibili da macchine. Inoltre non ritengo fondato porre il confine fra "intelligenza" umana ed automatica nella statistica e/o probabilità perchè anche noi - convinzione mia - possiamo essere definiti macchine statistiche, ed il nostro apprendimento deve molto alla statistica. Vedo un futuro promettente [2] oppure inquietante soprattutto in conseguenza della nostra scelta di imparare ad usare attivamente la macchina oppure di subirla passivamente, finchè pochissimi sanno usarla sono molto pessimista. Cordialmente. Duccio (Alessandro Marzocchi) [1] Griffin, The question of animal awareness - https://books.rupress.org/catalog/book/question-animal-awareness [2] https://www.infodata.ilsole24ore.com/2025/06/15/il-divario-tra-uomini-e-cha…

1 0

LLMs in competitive programming
by Daniela Tafani June 17, 2025

June 17, 2025

LiveCodeBench Pro: How Do Olympiad Medalists Judge LLMs in Competitive Programming? [Submitted on 13 Jun 2025] Abstract Recent reports claim that large language models (LLMs) now outperform elite humans in competitive programming. Drawing on knowledge from a group of medalists in international algorithmic contests, we revisit this claim, examining how LLMs differ from human experts and where limitations still remain. We introduce LiveCodeBench Pro, a benchmark composed of problems from Codeforces, ICPC, and IOI that are continuously updated to reduce the likelihood of data contamination. A team of Olympiad medalists annotates every problem for algorithmic categories and conducts a line-by-line analysis of failed model-generated submissions. Using this new data and benchmark, we find that frontier models still have significant limitations: without external tools, - the best model achieves only 53% pass@1 on medium-difficulty problems - and 0% on hard problems, domains where expert humans still excel. We also find that LLMs succeed at implementation-heavy problems but struggle with nuanced algorithmic reasoning and complex case analysis, often generating confidently incorrect justifications. High performance appears largely driven by implementation precision and tool augmentation, not superior reasoning. LiveCodeBench Pro thus highlights the significant gap to human grandmaster levels, while offering fine-grained diagnostics to steer future improvements in code-centric LLM reasoning. <https://livecodebenchpro.com/> <https://arxiv.org/abs/2506.11928>

1 0

[RAI] Building an Ecosystem for Responsible, Trustworthy Governance of AI Solutions
by Daniele Quercia June 17, 2025

June 17, 2025

Title: Building an Ecosystem for Responsible, Trustworthy Governance of AI Solutions Elizabeth M. Daly, IBM Research Join the meeting <https://teams.microsoft.com/l/meetup-join/19%3ameeting_ZmFlYWRhYWUtNmY3Zi00…> Abstract: As large language models (LLMs) rapidly reshape the technological landscape, their potential is matched by profound risks. From accelerating innovation and development to the growing reliance on AI in high-stakes decision-making, we face a critical challenge: how do we ensure these systems are aligned with human values, factually reliable, and governed responsibly? This talk explores the urgent need for a governance-first ecosystem that empowers developers, researchers, and policymakers to evaluate, align, and manage AI systems effectively. We present a community-driven ecosystem of tools: EvalAssist, Risk Atlas Nexus, FactReasoner, Benchmark Cards, and Unitxt—designed to support transparent evaluation, risk-aware development, and factual reliability. By turning risks into concrete evaluations and mitigations, we pave the way for AI systems that are as responsible as they are innovative. <https://arxiv.org/abs/2407.02191>Bio: Elizabeth M. Daly is a Senior Technical Staff Member leading the Interactive AI Group at IBM Research in Dublin. Her work focuses on Human-Centered Trustworthy AI, developing systems that align with human values and support collaborative decision-making. She earned her Ph.D. in Computer Science from Trinity College Dublin, where her research explored social network analysis in delay-tolerant networks. Before joining IBM Research Dublin, she was a postdoctoral researcher at IBM’s Research Center for Social Software in Cambridge, MA. Elizabeth has served as a Master Inventor at IBM, was vice-chair the Royal Irish Academy’s Committee on Engineering and Computer Science, and serves on the program committees of leading conferences including AAAI, IUI, RecSys, WWW, UMAP, and ICWSM. Subscribe to future talk announcements: Anyone outside Bell Labs can receive talk announcements by subscribing to the mailing list. To subscribe, send an empty email with the subject line "Subscribe RAI" to daniele.quercia(a)polito.it<mailto:daniele.quercia@polito.it>

1 0

123° Nexa Lunch Seminar e 184° Mercoledì di Nexa | 25 giugno 2025
by Nexa - Media June 17, 2025

June 17, 2025

Gentilissime, gentilissimi, Vi invitiamo a partecipare a due incontri che si terranno nella giornata di mercoledì 25 giugno: * 123° Nexa Lunch Seminar, dal titolo “L’estetica dell’invisibilità nell’era digitale: privacy, anonimato e deepfakes” Ospite: Remo Gramigna (Università di Torino) Dalle ore 13.00 alle ore 14.00 Maggiori dettagli: https://nexa.polito.it/lunch-123 * 184° Mercoledì di Nexa, dal titolo “A Law and Political Economy of Intellectual Property" Ospite: Talha Syed (UC Berkeley School of Law) Dalle ore 17.00 alle ore 19.00 Maggiori dettagli: https://nexa.polito.it/mercoledi-184 Entrambi gli incontri si terranno IN PRESENZA e ONLINE. SEDE FISICA 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 degli incontri: https://didattica.polito.it/VClass/NexaEvent 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

Re: [nexa] The Problem With "AI Fluency"
by alessandro marzocchi June 14, 2025

June 14, 2025

Quale fluency per gli analfabeti? Quanti sanno leggere e scrivere il codice digitale? Da pochi decenni - e dopo qualche migliaio di anni - abbiamo quasi sconfitto l'analfabetismo ed ora ce lo ritroviamo. Buona domenica, cordialmente, Duccio (Alessandro Marzocchi) Il giorno sab 14 giu 2025 alle ore 12:00 <nexa-request(a)server-nexa.polito.it> ha scritto: > Message: 1 > Date: Fri, 13 Jun 2025 15:44:31 +0000 > From: Daniela Tafani <daniela.tafani(a)unipi.it> > To: alessandro marzocchi <alemarzoc(a)gmail.com>, nexa > <nexa(a)server-nexa.polito.it> > Subject: Re: [nexa] The Problem With "AI Fluency" > > Buon pomeriggio, Alessandro > > "AI Fluency", a me pare lo stesso argomento che chiamai > "alfabetizzazione digitale". > No, non lo è. >

1 0

Re: [nexa] The Problem With "AI Fluency"
by alessandro marzocchi June 13, 2025

June 13, 2025

"AI Fluency", a me pare lo stesso argomento che chiamai "alfabetizzazione digitale". Continuo a ritenere che il digitale avvia un cambiamento "epocale", cronologicamente secondo a quello della parola, cioè stiamo passando dall'epoca del codice alfabetico all'epoca del codice digitale. In proposito ricordo le critiche, fondate sulla convinzione che il digitale, ed in particolare AI, si basa su statistica e probabilità, diversamente da noi umani. ... diversamente da noi umani, questa è la convinzione diffusa, ma domando: *davvero il nostro apprendimento non è statistico, probabilistico?* Cordialmente, Duccio (Alessandro Marzocchi) Il giorno mar 10 giu 2025 alle ore 11:48 <nexa-request(a)server-nexa.polito.it> ha scritto: > Date: Mon, 9 Jun 2025 17:40:30 +0000 > From: Daniela Tafani <daniela.tafani(a)unipi.it> > Subject: [nexa] The Problem With "AI Fluency" > 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> > ------------------------------ > >

2 2

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