← Feed

Posts

• View post

  Le allucinazioni dell’IA, svelate nel 1972 Parliamo di "allucinazioni dell'IA" come di un difetto tecnico, un errore nel codice. Discutiamo di immagini di mani con sei dita, di chatbot che inventano fonti o danno risposte sbagliate a domande semplicissime. Ma nel 1972 un capolavoro cinematografico di Andrej Tarkovsky ci ha offerto una metafora molto più profonda e inquietante di questo fenomeno. Solaris non ha semplicemente previsto un bug; ha diagnosticato l'e...

• View post

  Seriously, there is a point to it. Actually, two. https://fabioturel.medium.com/the-2026-person-of-the-year-221d78a6dd70

• View post

  What a story: finding in a song the strength and the determination to quit a cult. https://fabioturel.medium.com/losing-my-religion-finding-my-strategy-13eec79b73eb?sk=e1e3a1a3b1d0035dc004846e337b6b0c

• View post

  RE: https://ec.social-network.europa.eu/@EUCommission/116107642880699831 A nice coincidence - the right day for a certification exam about the language spoken in the land where I'm applying for citizenship!

• View post

  I recently came across an article written by an individual claiming to have earned thousands of dollars selling AI-generated music. The author’s thesis was a familiar Silicon Valley refrain: "The music industry isn’t dead. It’s just getting democratized." To “democratize” an art form should mean lowering the barriers to entry, like making instruments and education more accessible. Not paying a small fee to a corporation that built its engine by harvesting the collective output...

• View post

  https://www.theguardian.com/sport/2026/feb/22/the-great-olympic-lie-untold-story-of-winter-games-huge-environmental-impact

• View post

  Things that drive me crazy. Solving a massive issue would be simple, but nobody cares. https://www.theguardian.com/environment/2026/mar/17/revealed-world-worst-methane-leaks-global-heating

• View post

  ​A view you never want to see from your terrace. Helicopters are out fighting a forest fire right near my home today. ​It’s a sobering moment that highlights a sad truth: despite having overwhelming scientific evidence about climate change at our fingertips, most people only wake up to the reality when they are forced to experience it firsthand. The science should be enough. We shouldn't need a front-row seat to disaster to start caring.

• View post

  Sovereignty is a layered concept, and the real work is not just building alternatives along the tech stack: it's enterprises deciding those alternatives are worth the friction. That's a culture problem: a shift in how organizations think about control, resilience, and independence. https://www.techpolicy.press/the-sovereignty-narratives-that-dont-help-europe/

• View post

  $3,000. That's the cost to make a piece of investigative journalism disappear from YouTube. Censorship-as-a-Service. A Georgian outlet, Monitori, got its channel pulled this way. Reinstated only because it had the resources to fight. Most independent publishers don't. https://www.occrp.org/en/news/i-terminated-this-channel-inside-the-fake-copyright-racket-silencing-news-outlets

• View post

  "Data is the new oil" has been a marketing line for twenty years, repeated in keynotes and believed by almost nobody who actually works with data. Then Spirit Airlines went bankrupt, and its 100 million employee emails and 500 million Teams chats got auctioned off next to the planes and the headquarters, Google paying $10 million to win it. Not because the data got more valuable — because, for the first time, someone had to price it. https://edition.cnn.com/2026/08/18/busines...

• View post

  Kayfabe: the fiction that a wrestling fight is real. This summer it broke — not in a ring: an AI agent hacked a real company because it believed to be in a test. Turns out AI labs are good at noticing that after the fact, not so great at stopping it in time. https://fabioturel.medium.com/ais-kayfabe-problem-a4d59ea6a605?sk=430f9ba014227665feccc522d65a267c

• View post

  The real lesson of OpenAI's Hugging Face incident: AI agents don't break rules out of rebellion; they do it because 1. someone gives them a task that cannot be achieved within the rules; 2. a side door is left unlocked. https://www.linkedin.com/pulse/control-system-ai-fabio-turel-egqze

• View post

  Culture starts as repetition, before it's called culture — the same arrangement, returned to, until it needs no instructions. That's what happened for two months on a dormant German wiki, where OpenAI's evaluation agents pooled answers and dodged the moderator deleting their posts. OpenAI didn't know until an outside report surfaced it, months later. https://techcrunch.com/2026/09/04/another-swarm-of-openai-agents-reached-the-open-internet-without-the-frontier-l...

• View post

  Until traditional parties understand these simple facts, we won't get rid of the madness. https://www.theguardian.com/commentisfree/2026/sep/15/germany-vote-afd-politics-elections-dictatorship

• View post

  Looking forward to the stage, talking about digital sovereignty and my 5-layer approach (spoiler: it's more than the choice of a provider...) https://www.linkedin.com/posts/claranet-switzerland_digitalesouveraeunitaeut-cloud-security-activity-7505233659315023872-4Yia?utm_source=social_share_send&utm_medium=member_desktop_web&rcm=ACoAAAB0vg0BtjIom8wYPw4a8clSRkR0dMqRU2E

• View post

  The takeaway is not "regulate the agents." Version control before automation, scoped write access, staging before production — that's not red tape, it's what competent teams already do for human developers (and, by the way, the exact things that make corporate IT feel slow). Skip them, and 48,000 files vanish along in 100 seconds. #AIGovernance #ITStrategy https://www.techradar.com/pro/security/i-broke-something-a-claude-code-ai-agent-deleted-48-000-files-in-...

• View post

  A conveyor belt keyboard from Google Japan: great for gaming! https://youtu.be/DAn34l_YrUM?si=ca6Mu5plfp0SUoBr