- nexa - server-nexa.polito.it

Re: [nexa] Perché Richard Stallman sbaglia in tema di intelligenza artificiale
by Guido Vetere Feb. 14, 2025

Feb. 14, 2025

Beppe, il medico che fa una diagnosi per analogia rispetto ai casi che ha osservato, è comunque in grado di spiegare il suo ragionamento sulla base di una teoria. E anche se riproduce 'a orecchio' migliaia di dimostrazioni osservate, è (o dovrebbe essere) in grado di fornire la spiegazione sulla base di assiomi e regole di inferenza. Questo è il patto sociale quando andiamo a farci curare da un laureato in medicina piuttosto che da uno sciamano. La fisica del cervello del medico, e la sua analogia con un LLM, non c'entra nulla: la teoria medica vale in quanto 'oggetto sociale', non come 'campo di forze'. G. On Fri, 14 Feb 2025 at 09:13, Giuseppe Attardi <giuseppe.attardi(a)unipi.it> wrote: > > > On 14 Feb 2025, at 07:47, Guido Vetere <vetere.guido(a)gmail.com> wrote: > > > Beppe, > ma la "spiegazione" della c.d. chain-of-thought si trova sullo stesso > piano epistemico di ciò che intende spiegare, cioè quello della > correlazione, non quello delle causalità. > La differenza è 'striking' e la spiega bene Judea Pearl nel suo "The Book > of Why" (https://en.wikipedia.org/wiki/The_Book_of_Why) > > Certo, perché tu pensi che quando un medico ti spiega la ragione della sua > diagnosi non faccia lo stesso? O un matematico ti presenta una > dimostrazione senza accennare all’intuizione che l’ha portato a concepirla? > Ricordi Ramanujan? > Sono tutte ricostruzioni razionali a posteriori di un processo largamente > inconscio. > > Se pensiamo che l’intelligenza sia solo seguire pedissequamente un > procedimento algoritmico, o l’applicazione meccanica di regole logiche, > abbiamo una scarsa concezione dell’intelligenza. > Del resto sbbiamo provato per 50 a riprodurla pensando che fosse così e > abbiamo fallito. > > — > > G. > > On Thu, 13 Feb 2025 at 12:36, Giuseppe Attardi <attardi(a)di.unipi.it> > wrote: > >> Non solo lui, ma gran parte degli studiosi di linguistica della vecchia >> generazione, a partire da Noam Chomsky, sono rimasti indietro di 8 anni. >> Non solo, ma non comprendono la differenza tra i LLM e i chatbot, che >> sono delle applicazioni dei primi, nate inizialmente per gioco: ricorderete >> la storia degli unicorni, prodotta da GPT-2. >> Era un esercizio classico di uso dei LM per generare testo a >> completamento di un prompt. >> >> Ma I chatbot sono un’applicazione specializzata dei LLM, allenata con una >> fase di post-training, con varie tecniche, in primis il RLHF introdotto in >> ChatGPT, per addestrarlo a partecipare a dialoghi, ossia ad accontentare >> gli interlocutori. >> Ma oltre ai chatbot, ci sono mille altre applicazioni dei LLM che non >> sono solo per chiacchierare. >> >> Da allora, la tecnica si è poi ulteriormente evoluta con tre sostanziali >> progressi: >> >> 1. Con l’aumentare della scala dei modelli, sono apparse capacità >> emergenti (emergent abilities), che vanno oltre la banale capacità di >> predire la prossima parola: un fenomeno che si spiega con la teoria dei >> sitemi complessi di Giorgio Parisi: l’applicazione su larga scala di >> semplici funzioni di probabilità dà origine a comportamenti complessi, non >> riducibili alla funzione di partenza >> 2. Si sono raffinate le tecniche di post-processing: SFT e RL basato su >> DPO (Direct Preference Optimizazion) o GRPO (quella usata da DeepSeek R1 >> ecc.) Quest’ultima tecnica accelera l’apprendimento con RL e viene usata >> per insegnare direttamente a effettuare ragionamenti matematici e logici ai >> modelli, senza bisogno di un secondo modello di critica delle risposte come >> in ChatGPT. >> 3. Le capacità apprese dai modelli di grandissime dimensioni possono >> essere “distillate” in modelli più piccoli, mantenendone le capacità >> acquisite. >> >> Quindi i modelli attuali, come GPT-4 o3, DeepSeek R1, Gemini 2.0, ecc., >> fanno cose ben diverse dalla semplice generazione a caso di risposte. >> DeepSeek è particolarmente interessante da osservare, perché riporta >> nella risposta tutte le fasi del suo ragionamento, racchiuse tra i tag >> <think></think>, mentre gli altri modelli li tengono nascosti. >> Si vede chiaramente come svolge il suo ragionamento: propone una prima >> risposta, poi la valuta criticamente, dicendo: “Ah wait. …” e spiegando >> come quella risposta funziona e se ci sono criticità, poi ne genera una >> seconda che risolve quelle criticità e poi ci ragiona sopra di nuovo. >> >> Questo purtroppo in Italia ci è vietato dalla decisione del Garante della >> Privacy che ci ha impedito l’accesso a DeepSeek. >> >> Ma è un passo avanti importante, anche perché rintuzza un’altra critica >> ai modelli ML, la mancanza di trasparenza. >> In questo caso, l’intero processo di ragionamento viene esposto, compresa >> una spiegazione in termini perfettamente comprensibili della ragione della >> risposta. >> >> — Beppe >> >> >> On 13 Feb 2025, at 12:00, nexa-request(a)server-nexa.polito.it wrote: >> >> From: Diego Giorio <dgiorio(a)hotmail.com> >> To: Nexa <nexa(a)server-nexa.polito.it> >> Subject: [nexa] Perché Richard Stallman sbaglia in tema di >> intelligenza artificiale >> Message-ID: >> < >> BN6PR17MB3139F372CA9F7422D383438FBEFF2(a)BN6PR17MB3139.namprd17.prod.outlook.com >> > >> >> Content-Type: text/plain; charset="iso-8859-1" >> >> Ieri è stata una bellissima esperienza. >> >> A titolo personale mi pongo un po' a metà tra l'opinione di Stallman e >> quella di questo articolo, che comunque ritengo giusto segnalare >> >> Buona giornata a tutti >> >> >>

3 2

DeepSeek fa sballare l’AI Act
by Giuseppe Attardi Feb. 14, 2025

Feb. 14, 2025

DeepSeek infligge un duro colpo anche all'European AI Act, che stabilisce una soglia per l'applicazione delle sue regole in base a un numero fisso di FLOPS utilizzati nell'addestramento di un modello: 10^25. Quando questa soglia viene raggiunta, si presumerà che un modello sia un modello GPAI che presenta rischi sistemici. Una costante universale scolpita nella pietra dalla legge (anche se ammettono che potrebbero doverla rivedere): come fissare per legge la velocità della luce. DeepSeek V3 è effettivamente al di sotto di quella soglia: 3.29e24 FLOP nell'addestramento —

1 0

Evidence For (and Against) Emergent World Models in LLM
by Daniela Tafani Feb. 14, 2025

Feb. 14, 2025

Melanie Mitchell, LLMs and World Models, Part 2 Evidence For (and Against) Emergent World Models in LLM Feb 13, 2025 [...] Conclusion The claims of emergent abstract world models in LLMs are not yet supported by strong evidence. There is some evidence of such world models arising in transformers trained on narrow domains (Othello, chess, mazes, etc.) but also evidence that their abilities arise not from human-like internal models but from large “bags of heuristics”. Moreover, the notion of “world model” itself is not rigorously defined; when considering whether an agent has a particular kind of world model, we should ask what kinds of questions such a model should be able to answer, how easy or hard it should be for the agent to get answers from the model, and to what extent we would expect that the model would allow the agent to adapt to novel situations. <https://aiguide.substack.com/p/llms-and-world-models-part-2> Qui la prima parte <https://aiguide.substack.com/p/llms-and-world-models-part-1>

1 0

Cory Doctorow, Premature Internet Activists
by Daniela Tafani Feb. 14, 2025

Feb. 14, 2025

Cory Doctorow, Premature Internet Activists Posted on February 13, 2025 "Premature antifacist" was a sarcastic term used by leftists caught up in the Red Scare to describe themselves, as they came under ideological suspicion for having traveled to Spain to fight against Franco's fascists before the US entered WWII and declared war against the business-friendly, anticommunist fascist Axis powers of Italy, Spain, Greece, and, of course, Germany: https://www.google.com/books/edition/In_Denial/fBSbKS1FlegC?hl=en&gbpv=… The joke was that opposing fascism made you an enemy of America – unless you did so after the rest of America had woken up to the existential threat of a global fascist takeover. What's more, if you were a "premature antifascist," you got no credit for fighting fascism early on. Quite the contrary: fighting fascism before the rest of the US caught up with you didn't make you prescient – it made you a pariah. I've been thinking a lot about premature antifascism these days, as literal fascists use the internet to coordinate a global authoritarian takeover that represents an existential threat to a habitable planet and human thriving. In light of that, it's hard to argue that the internet is politically irrelevant, and that fights over the regulation, governance, and structure of the internet are somehow unserious. And yet, it wasn't very long ago that tech policy was widely derided as a frivolous pursuit, and that tech organizing was dismissed as "slacktivism": https://www.newyorker.com/magazine/2010/10/04/small-change-malcolm-gladwell Elevating concerns about the internet's destiny to the level of human rights struggle was delusional, a glorified argument about the rules for forums where sad nerds argued about Star Trek. If you worried that Napster-era copyright battles would make it easy to remove online content by claiming that it infringed copyright, you were just carrying water for music pirates. If you thought that legalizing and universalizing encryption technology would safeguard human rights, you were a fool who had no idea that real human rights battles involved confronting Bull Connor in the streets, not suing the NSA in a federal courtroom. And now here we are. Congress has failed to update consumer privacy law since 1988 (when they banned video store clerks from blabbing about your VHS rentals). Mass surveillance enables everything from ransomware, pig butchering and identity theft to state surveillance of "domestic enemies," from trans people to immigrants. What's more, the commercial and state surveillance apparatus are, in fact, as single institution: states protect corporations from privacy law so that corporations can create and maintain population-scale nonconsensual dossiers on all the intimate facts of our lives, which governments raid at will, treating them as an off-the-books surveillance dragnet: https://pluralistic.net/2023/08/16/the-second-best-time-is-now/#the-point-o… Our speech forums have been captured by billionaires who censor anti-oligarchic political speech, and who spy on dissident users in order to aid in political repression. Bogus copyright claims are used to remove or suppress disfavorable news reports of elite rapists, thieves, war criminals and murderers: https://pluralistic.net/2024/06/27/nuke-first/#ask-questions-never You'd be hard pressed to find someone who'd describe the fights over tech governance in 2025 as frivolous or disconnected from "real politics" This is where the premature antifascist stuff comes in. An emerging revisionist history of internet activism would have you believe that the first generation of tech liberation activists weren't fighting for a free, open internet – we were just shilling for tech companies. The P2P wars weren't about speech, privacy and decentralization – they were just a way to help the tech sector fight the entertainment industry. DRM fights weren't about preserving your right to repair, to privacy, and to accessibility – they were just about making it easy to upload movies to Kazaa. Fighting for universal access to encryption wasn't about defending everyday people from corporate and state surveillance – it was just a way to help terrorists and child abusers stay out of sight of cops. Of course, now these fights are all about real things. Now we need to worry about centralization, interoperability, lock-in, surveillance, speech, and repair. But the people – like me – who've been fighting over this stuff for a quarter-century? We've gone from "unserious fools who mistook tech battles for human rights fights" to "useful idiots for tech companies" in an eyeblink. "Premature Internet Activists," in other words. This isn't merely ironic or frustrating – it's dangerous. Approaching tech activism without a historical foundation can lead people badly astray. For example, many modern tech critics think that Section 230 of the Communications Decency Act (which makes internet users liable for illegal speech acts, while immunizing entities that host that speech) is a "giveaway to Big Tech" and want to see it abolished. Boy is this dangerous. CDA 230 is necessary for anyone who wants to offer a place for people to meet and discuss anything. Without CDA 230, no one could safely host a Mastodon server, or set up the long-elusive federated Bluesky servers. Hell, you couldn't even host a group-chat or message board: https://www.techdirt.com/2020/06/23/hello-youve-been-referred-here-because-… Getting rid of CDA 230 won't get rid of Facebook or make it clean up its act. It will just make it impossible for anyone to offer an alternative to Facebook, permanently enshrining Zuck's dominance over our digital future. That's why Mark Zuckerberg wants to kill Section 230: https://www.nbcnews.com/tech/tech-news/zuckerberg-calls-changes-techs-secti… Defending policies that make it easier to host speech isn't the same thing as defending tech companies' profits, though these do sometimes overlap. When tech platforms have their users' back – even for self-serving reasons – they create legal precedents and strong norms that protect everyone. Like when Apple stood up to the FBI on refusing to break its encryption: https://en.wikipedia.org/wiki/Apple%E2%80%93FBI_encryption_dispute If Apple had caved on that one, it would be far harder for, say, Signal to stand up to demands that it weaken its privacy guarantees. I'm no fan of Apple, and I would never mistake Tim Cook – who owes his CEOhood to his role in moving Apple production to Chinese sweatshops that are so brutal they had to install suicide nets – for a human rights defender. But I cheered on Apple in its fight against the FBI, and I will cheer them again, if they stand up to the UK government's demand to break their encryption: https://www.bbc.com/news/articles/c20g288yldko This doesn't make me a shill for Apple. I don't care if Apple makes or loses another dime. I care about Apple's users and their privacy. That's why I criticize Apple when they compromise their users' privacy for profit: https://pluralistic.net/2024/01/12/youre-holding-it-wrong/#if-dishwashers-w… The same goes for fights over scraping. I hate AI companies as much as anyone, but boy is it a mistake to support calls to ban scraping in the name of fighting AI: https://pluralistic.net/2023/09/17/how-to-think-about-scraping/ It's scraping that lets us track paid political disinformation on Facebook (Facebook isn't going to tell us about it): https://pluralistic.net/2021/08/05/comprehensive-sex-ed/#quis-custodiet-ips… And it's scraping that let us rescue all the CDC and NIH data that Musk's broccoli-hair brownshirts deleted on behalf of DOGE: https://www.cnet.com/tech/services-and-software/how-to-access-important-hea… It's such a huge mistake to assume that anything corporations want is bad for the internet. There are many times when commercial interests dovetail with online human rights. That's not a defense of capitalism, it's a critique of capitalism that acknowledges that profits do sometimes coincide with the public interest, an argument that Marx and Engels devote Chapter One of The Communist Manifesto to: https://www.nytimes.com/2022/10/31/books/review/a-spectre-haunting-china-mi… In the early 1990s, Al Gore led the "National Information Infrastructure" hearings, better known as the "Information Superhighway" hearings. Gore's objective was to transfer control over the internet from the military to civilian institutions. It's true that these institutions were largely (but not exclusively) commercial entities seeking to make a buck on the internet. It's also true that much of that transfer could have been to public institutions rather than private hands. But I've lately – and repeatedly – heard this moment described (by my fellow leftists) as the "privatization" of the internet. This is strictly true, but it's even more true to say that it was the demilitarization of the internet. In other words, corporations didn't take over functions performed by, say, the FCC – they took over from the Pentagon. Leftists have no business pining for the days when the internet was controlled by the Department of Defense. Caring about the technological dimension of human rights 30 years ago – or hell, 40 years ago – doesn't make you a corporate stooge who wanted to launch a thousand investment bubbles. It makes you someone who understood, from the start, that digital rights are human rights, that cyberspace would inevitably evert into meatspace, and that the rules, norms and infrastructure we built for the net would someday be as consequential as any other political decision. I'm proud to be a Premature Internet Activist. I just celebrated my 23rd year with the Electronic Frontier Foundation, and yesterday, we sued Elon Musk and DOGE: https://www.eff.org/press/releases/eff-sues-opm-doge-and-musk-endangering-p… <https://pluralistic.net/2025/02/13/digital-rights/#are-human-rights>

2 1

Re: [nexa] Perché Richard Stallman sbaglia in tema di intelligenza artificiale
by Giuseppe Attardi Feb. 14, 2025

Feb. 14, 2025

Non solo lui, ma gran parte degli studiosi di linguistica della vecchia generazione, a partire da Noam Chomsky, sono rimasti indietro di 8 anni. Non solo, ma non comprendono la differenza tra i LLM e i chatbot, che sono delle applicazioni dei primi, nate inizialmente per gioco: ricorderete la storia degli unicorni, prodotta da GPT-2. Era un esercizio classico di uso dei LM per generare testo a completamento di un prompt. Ma I chatbot sono un’applicazione specializzata dei LLM, allenata con una fase di post-training, con varie tecniche, in primis il RLHF introdotto in ChatGPT, per addestrarlo a partecipare a dialoghi, ossia ad accontentare gli interlocutori. Ma oltre ai chatbot, ci sono mille altre applicazioni dei LLM che non sono solo per chiacchierare. Da allora, la tecnica si è poi ulteriormente evoluta con tre sostanziali progressi: 1. Con l’aumentare della scala dei modelli, sono apparse capacità emergenti (emergent abilities), che vanno oltre la banale capacità di predire la prossima parola: un fenomeno che si spiega con la teoria dei sitemi complessi di Giorgio Parisi: l’applicazione su larga scala di semplici funzioni di probabilità dà origine a comportamenti complessi, non riducibili alla funzione di partenza 2. Si sono raffinate le tecniche di post-processing: SFT e RL basato su DPO (Direct Preference Optimizazion) o GRPO (quella usata da DeepSeek R1 ecc.) Quest’ultima tecnica accelera l’apprendimento con RL e viene usata per insegnare direttamente a effettuare ragionamenti matematici e logici ai modelli, senza bisogno di un secondo modello di critica delle risposte come in ChatGPT. 3. Le capacità apprese dai modelli di grandissime dimensioni possono essere “distillate” in modelli più piccoli, mantenendone le capacità acquisite. Quindi i modelli attuali, come GPT-4 o3, DeepSeek R1, Gemini 2.0, ecc., fanno cose ben diverse dalla semplice generazione a caso di risposte. DeepSeek è particolarmente interessante da osservare, perché riporta nella risposta tutte le fasi del suo ragionamento, racchiuse tra i tag <think></think>, mentre gli altri modelli li tengono nascosti. Si vede chiaramente come svolge il suo ragionamento: propone una prima risposta, poi la valuta criticamente, dicendo: “Ah wait. …” e spiegando come quella risposta funziona e se ci sono criticità, poi ne genera una seconda che risolve quelle criticità e poi ci ragiona sopra di nuovo. Questo purtroppo in Italia ci è vietato dalla decisione del Garante della Privacy che ci ha impedito l’accesso a DeepSeek. Ma è un passo avanti importante, anche perché rintuzza un’altra critica ai modelli ML, la mancanza di trasparenza. In questo caso, l’intero processo di ragionamento viene esposto, compresa una spiegazione in termini perfettamente comprensibili della ragione della risposta. — Beppe > On 13 Feb 2025, at 12:00, nexa-request(a)server-nexa.polito.it wrote: > > From: Diego Giorio <dgiorio(a)hotmail.com <mailto:dgiorio@hotmail.com>> > To: Nexa <nexa(a)server-nexa.polito.it <mailto:nexa@server-nexa.polito.it>> > Subject: [nexa] Perché Richard Stallman sbaglia in tema di > intelligenza artificiale > Message-ID: > <BN6PR17MB3139F372CA9F7422D383438FBEFF2(a)BN6PR17MB3139.namprd17.prod.outlook.com <mailto:BN6PR17MB3139F372CA9F7422D383438FBEFF2@BN6PR17MB3139.namprd17.prod.outlook.com>> > > Content-Type: text/plain; charset="iso-8859-1" > > Ieri è stata una bellissima esperienza. > > A titolo personale mi pongo un po' a metà tra l'opinione di Stallman e quella di questo articolo, che comunque ritengo giusto segnalare > > Buona giornata a tutti

7 7

Re: [nexa] Perché Richard Stallman sbaglia in tema di intelligenza artificiale
by Giuseppe Attardi Feb. 14, 2025

Feb. 14, 2025

> On 13 Feb 2025, at 13:01, Fabio Alemagna <falemagn(a)gmail.com> wrote: >> [...] >> Questo purtroppo in Italia ci è vietato dalla decisione del Garante della Privacy che ci ha impedito l’accesso a DeepSeek. > > Lo si può comunque scaricare sul proprio computer e farlo girare in locale. Qui un tutorial: https://www.datacamp.com/tutorial/deepseek-r1-ollama Sto sperimentando coi modelli DeepSeek da parecchio tempo, prima che diventassero famosi e non è così semplice. Il modello DeepSeek V3 ha 617B parameters, ossia occupa 1.2TB di memoria: solo per caricarlo ci vogliono 8 GPU da 80GB. Lo dicono loro stessi, nel technical report, che il modello è pesante to deploy e richiederà ulteriori studi. Si possono usare modelli più piccoli, ma le prestazioni calano e comunque non sono velocissimi. Il modello DeepSeek-Coder-33-instruct impiega circa 1 minuto a risposta. DeepSeek lascia a terze parti di offrire accesso online, perché non ce la fanno. Shameless push. Non avendo grandi risorse, siamo partiti dal più piccolo dei modelli e abbiamo fatto SFT e RL con Preference Optimization sulla challenge Semeval 2025 Tabular Question Answering. Ci siamo classificati secondi nella categoria modelli piccoli. > > Fabio

1 0

Pappagalli stocazzici (pun intended)
by Fabio Alemagna Feb. 14, 2025

Feb. 14, 2025

*Language Models Use Trigonometry to Do Addition* Subhash Kantamneni, Max Tegmark MIT 2025 https://arxiv.org/abs/2502.00873 <https://arxiv.org/abs/2502.00873?fbclid=IwZXh0bgNhZW0CMTAAAR3eeuot_BGe0CySv…> "We first discover that numbers are represented in these LLMs as a generalized helix, which is strongly causally implicated for the tasks of addition and subtraction, and is also causally relevant for integer division, multiplication, and modular arithmetic. We then propose that LLMs compute addition by manipulating this generalized helix using the “Clock” algorithm: to solve a + b, the helices for a and b are manipulated to produce the a + b answer helix which is then read out to model logits. " Imagine teaching a computer to do math, not by giving it step-by-step rules like a calculator, but by letting it figure things out on its own. Scientists have been trying to understand how large language models (LLMs)—the AI behind tools like ChatGPT—actually "think" when solving even basic math problems. In this study, researchers uncovered something fascinating: LLMs seem to use trigonometry to perform addition. Instead of treating numbers as simple values, these AI models represent them as points on a twisted spiral—like a helix. When the model needs to add two numbers, it doesn’t just line them up and sum them. Instead, it rotates and shifts these spiral representations using what the researchers call the "Clock" algorithm. Essentially, the AI aligns the spirals for two numbers in a way that lands it on the correct result. By tracking how different parts of the AI process these helical number patterns—from individual neurons to larger structures called attention heads—scientists confirmed that this method is not just a coincidence but a fundamental part of how the model handles math. This discovery provides the first detailed glimpse into how AI "sees" numbers and could help us design smarter, more reliable models in the future. ___ via Cecile Tamura: https://www.facebook.com/cecile.tamura/posts/pfbid0U7FwUxbXoB5CtPV3BgC7WoLc…

2 2

Intelligenze aliene
by Guido Vetere Feb. 14, 2025

Feb. 14, 2025

È uscito "Intelligenze aliene", un agile libretto dove cerco di connettere l'attuale vicenda dell'AI generativa alla sua genesi e a certe questioni molto aperte della filosofia del linguaggio novecentesca. https://lucasossellaeditore.it/libro/intelligenze-aliene/ Lo presenteremo a Roma il 25 Febbraio alle ore 18:30 presso la libreria Spazio Sette. Interverranno Maurizio Lenzerini (Sapienza) e Mario De Caro (Roma Tre). Modererà Luca De Biase. Chi è a Roma può prenotare il posto su eventbrite: https://www.eventbrite.it/e/biglietti-presentazione-di-intelligenze-aliene-… Siamo in attesa di sapere se ci sarà una diretta streaming, vi terrò informati. Spero di vedervi, o di sapere che siete lì :-) Guido

1 0

The CDC’s Website Is Being Actively Purged to Comply With Trump DEI Order
by Alberto Cammozzo Feb. 14, 2025

Feb. 14, 2025

<https://www.404media.co/the-cdcs-website-is-being-actively-purged-to-comply…> “CDC’s website is being modified to comply with President Trump’s Executive Orders.“ Large parts of the CDC’s website and several important databases were taken down on Friday and Saturday to comply with Trump’s executive orders banning DEI content. Saturday, a message at the top of the CDC’s home page said the website “is being modified to comply with President Trump’s Executive Orders.” CDC websites and databases taken offline include the CDC Atlas, the CDC Youth Risk Behavior Surveillance System, a CDC website about HIV treatment, and the CDC Social Vulnerability Index. Some of these removals were earlier reported by NBC News. Some of the pages were replaced with messages that read “Page Not Found or Temporarily Unavailable” or “The page you're looking for was not found.” There was widespread uncertainty throughout Friday as to whether a broader takedown across the government would happen. “Our team’s government affairs firm is advising that as of 5pm today, all U.S. government agency websites will be taken down,” an internal email obtained by 404 Media earlier Friday read. “According to reports, agencies are unable to comply fast enough with President Trump’s EO ordering all government entities to remove all DEI references from their websites, so these websites will be taken offline. There is no word on when they will be made available again.” At 5pm Friday, however, no widespread, cross-government takedowns happened. Throughout the day Friday and Saturday, however CDC pages continued to disappear. Saturday, a message at the top of the CDC’s website said “CDC’s website is being modified to comply with President Trump’s Executive Orders.” 404 Media has reported on U.S. government pages about gender identity were taken down; that GitHub commits showed the Trump administration scrubbing government web pages in real time; and how archivists are working to save thousands of datasets disappearing from Data.gov. Some federal contractors and federal employees spent much of Friday afternoon panicking about the deletions, and there was uncertainty about what would be taken offline and how widespread the takedowns would be. A CDC employee that 404 Media granted anonymity to speak about sensitive issues said that they were told by the Office of the Chief Information Security Officer of the Department of Health and Human Services that all employees were told they had to delete their preferred pronouns from their email signatures by 5 PM Friday. Agencies were also ordered to “review all agency programs, contracts, and grants, and terminate any that promote or inculcate gender ideology” and to “take down all outward facing media (websites, social media accounts, etc.) that inculcate or promote gender ideology,” with a deadline of 5 PM Eastern Friday. Agencies were forced to “send an email to all agency employees announcing that the agency will be complying with Defending Women and this guidance.” Agencies have been ordered to create a report within the next week that includes “a complete list of actions taken in response to this guidance.” The specific executive order is Trump’s “Defending Women from Gender Ideology Extremism and Restoring Biological Truth to the Federal Government (Defending Women).” A similar message was posted to Reddit earlier on Friday. “We are being told that the CDC website is scheduled to go down by EOD today. Please share this with your partners and encourage them, as well as you should plan to download any significant information,” it reads. There have been several efforts to archive data that already existed across the federal government, including the End of Term Archive, a volunteer effort that saved hundreds of terabytes of data before Trump was inaugurated.

1 0

Perché Richard Stallman sbaglia in tema di intelligenza artificiale
by Diego Giorio Feb. 13, 2025

Feb. 13, 2025

Ieri è stata una bellissima esperienza. A titolo personale mi pongo un po' a metà tra l'opinione di Stallman e quella di questo articolo, che comunque ritengo giusto segnalare Buona giornata a tutti https://www.ilsoftware.it/perche-richard-stallman-sbaglia-in-tema-di-intell…

2 1