July 2023 - nexa - server-nexa.polito.it

Artificial Intelligence: A New Frontier for Surveillance Capitalism?

by Federico Guerrini

"If you ask Alexa, Amazon’s voice assistant AI system, whether Amazon is a monopoly, it responds by saying it doesn’t know. It doesn’t take much to make it lambaste the other tech giants, but it’s silent about its own corporate parent’s misdeeds. When Alexa responds in this way, it’s obvious that it is putting its developer’s interests ahead of yours. Usually, though, it’s not so obvious whom an AI system is serving. To avoid being exploited by these systems, people will need to learn to approach AI skeptically. (...) As a security expert and data scientist, we believe that people who come to rely on these AIs will have to trust them implicitly to navigate daily life. That means they will need to be sure the AIs aren’t secretly working for someone else. Across the internet, devices and services that seem to work for you already secretly work against you. Smart TVs spy on you. Phone apps collect and sell your data. Many apps and websites manipulate you through dark patterns, design elements that deliberately mislead, coerce or deceive website visitors. This is surveillance capitalism, and AI is shaping up to be part of it." "Imagine asking your chatbot to plan your next vacation. Did it choose a particular airline or hotel chain or restaurant because it was the best for you or because its maker got a kickback from the businesses? As with paid results in Google search, newsfeed ads on Facebook and paid placements on Amazon queries, these paid influences are likely to get more surreptitious over time. If you’re asking your chatbot for political information, are the results skewed by the politics of the corporation that owns the chatbot? Or the candidate who paid it the most money? Or even the views of the demographic of the people whose data was used in training the model? Is your AI agent secretly a double agent? Right now, there is no way to know." https://theconversation.com/can-you-trust-ai-heres-why-you-shouldnt-209283 Ciao, Federico

July 23, 2023

The Santiago Boys

by J.C. DE MARTIN

E' online The Santiago Boys, il nuovo, splendido progetto di Evgeny Morozov: https://the-santiago-boys.com/ Qui sotto l'annuncio. Inoltre, oggi "La Lettura" dedica quattro pagine (!) all'iniziativa e ieri il Guardian da pubblicato un articolo a uno dei protagonisti della storia, Stafford Beer: https://www.theguardian.com/world/2023/jul/22/stafford-beer-chile-allende-t… Bravo, Evgeny! juan carlos / //The podcast that took 2+ years and 200+ interviews to produce is finally online!// // //You can listen to it on the main podcasting platforms (including Spotify and Apple Podcasts). // // //The website of the Santiago Boys offers plenty of extra materials for those of you who want to dig deeper: footnotes, backgrounders, sources, videos, a glossary, and so much else. // // //And we are also publishing the interviews with the many people we interviewed (check out, for example, this interview with Brian Eno where he talks about Stafford Beer and his own fascination with cybernetics). // // //Many thanks to dozens of people who worked on this ambitious project; you can see all their names here. // // //I hope you take a break from the Barbenheimer hyper and spend some time with the Santiago Boys instead! // // //Evgeny Morozov/

July 23, 2023

Re: [nexa] “Emergence” isn’t an explanation, it’s a prayer. A critique of Emergentism in Artificial Intelligence

by Giuseppe Attardi

Nei LLM, il concetto di "emergent ability” ha una definizione precisa: An ability that is “not present in small models but is present in large models.” https://www.jasonwei.net/blog/emergence Non mi pare che in questo contesto sia mai stata considerato di: "use it as a hypothesis to predict the outcome of a complex, unknown system, with the hope that a desired property will emerge;” Che il comportamento sorprendente e non facilmente spiegabile dei LLM, vada oltre le capacità per cui sono stati allenati, è un fatto appurato anche se controverso. Ne ho parlato con Giorgio Parisi, che sul tema dei sistemi complessi ha vinto il premio Nobel, e ha concordato con me che il fenomeno possa essere appunto spiegato come l’applicazione su larga scala di una semplice legge di probabilità: in questo caso la probabilità delle prossima parola in una sequenza. — Beppe > On 19 Jul 2023, at 10:37, nexa-request(a)server-nexa.polito.it wrote: > > Date: Wed, 19 Jul 2023 14:36:59 +0000 > From: Daniela Tafani <daniela.tafani(a)unipi.it <mailto:daniela.tafani@unipi.it>> > To: "nexa(a)server-nexa.polito.it <mailto:nexa@server-nexa.polito.it>" <nexa(a)server-nexa.polito.it <mailto:nexa@server-nexa.polito.it>> > Subject: [nexa] “Emergence” isn’t an explanation, it’s a > prayer. A critique of Emergentism in Artificial Intelligence > Message-ID: <8aced3bd15f24f548084f82dbf280719(a)unipi.it <mailto:8aced3bd15f24f548084f82dbf280719@unipi.it>> > Content-Type: text/plain; charset="windows-1252" > > “Emergence” isn’t an explanation, it’s a prayer > A critique of Emergentism in Artificial Intelligence > <https://ykulbashian.medium.com/?source=post_page-----ef239d3687bf----------…> > [cardboarddreams] > > Emergence is the notion that in a complex system, the interactions of the whole may exhibit properties that are not present in the individual parts. It is most often applied to examples in physics and nature, such as the collective behaviours of ant colonies, the self-organizing principles of social groups, or the macro properties of molecules. In the last few decades, it has given rise to emergentist perspectives of human cognition<https://onlinelibrary.wiley.com/doi/10.1111/j.1756-8765.2010.01116.x>, and even of consciousness<https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7597170/>. These are based on the recognition that the complexities and mysteries of the human mind, being a part of nature, may be characterized as emergent phenomena. > > This approach has an intuitive appeal. It is supported by the superficial facts: the brain — the source of intelligence and consciousness— is most certainly a complex of interconnected neurons. Emergentist interpretations of human behaviour may also boast some recent wins—the proliferation of LLMs (e.g. ChatGPT) may be seen as one such success. This has reignited the discussion of whether emergence is the best way to frame intelligence. > > You may have noticed that the last paragraph switched between two subtly different uses of “emergence”. The first use was to describe an observed emergent property; consciousness, we have seen it, likely emerges out of neuronal interactions. The second was to use it as a hypothesis to predict the outcome of a complex, unknown system, with the hope that a desired property will emerge; e.g. intelligence will arise from the interactions of artificial neurons at scale. The latter example is the driving motivation behind the multi-million dollar Human Brain Project<https://en.wikipedia.org/wiki/Human_Brain_Project>. The project justifies its cost by leaning on evidence from observed instances of emergence. Doing so paints emergence as a theory; that is, existing observations can be used to justify future predictions. > > But emergence is not a theory. Emergence can only be ascribed to a phenomenon in retrospect, once you already know what has “emerged”. The higher-level properties that emerge are qualitatively different from those at the lower-level — otherwise it wouldn’t be “emergence”. So by necessity they could not have been predicted from the lower-level ones. The properties of “intelligence” could not have been logically foreseen from the properties of neurons unless you had already observed that property emerge in a similar substrate. And even then it’s just a guess that is likely to be wrong given the complexity of the interactions involved; small differences can easily invalidate the hypothesis. In both cases emergence gives no new information: when explaining existing examples it gives you no new insights about the processes except that they happen; and when predicting unknown behaviours it gives very poor guarantees that anything you expect to happen will do so. > > Emergence is only really valid as a general metaphysical classification of certain phenomena. It’s a metaphysical category, like “cause”, “effect” or “change”. Using the word when explaining cognition is not wrong per se, it just has no real meaning or explanatory force. It’s like having a theory of “thing-happened-ness” — it’s correct, but void of content. Take, for example, the following quotes from a review article on emergence: > > This process gives rise to an emergent tendency to facilitate perception of items consistent with the patterns of English orthography, without explicitly representing this knowledge in a system of rules, as in other approaches. > > … > > However, such modes of thought themselves might be viewed as emergent consequences of a lifetime of thought-structuring practice supported by culture and education. > > Emergence in Cognitive Science, McClelland<https://onlinelibrary.wiley.com/doi/10.1111/j.1756-8765.2010.01116.x> > > If you removed the word “emergent” from the above two sentences, would anything important change? Indeed any sentence that includes “emergent” would give the same information if you removed it; “it gives rise to emergent properties” means the same as “it gives rise to properties”, or “there is an emergent tendency” is not substantially different from “there is a tendency”. > > Adding “emergent” to any sentence doesn’t increase its useful information content.¹ > > Emergence has no information that fundamentally differentiates it from a “miracle”. If I were to say that applying transformers<https://arxiv.org/abs/1706.03762> to Neural Networks creates intelligence through a miracle, I would be ridiculed. Were I to say that they create intelligence through emergent interactions, suddenly they gain an air of scientific credibility — but what have I added to the conversation with the use of that word? What quantifiable scientific facts are entailed in the term “emergent”? There are none. > > In cognitive science, emergence is regularly used to “explain” the connection between two phenomena, when it is otherwise complex and difficult to predict: e.g. how neuronal firing gives rise to consciousness, or transformers to the appearance of language comprehension. Where there may be a connection, but nothing more is known or can be proved, emergence is a placeholder that fills the gap. The word gives weight and gravitas to what is essentially a blank space. > > Despite emergence contributing nothing of substance to the discussion, as a concept it admittedly has a compelling intuitive appeal. There is a wonderful feeling about the notion of emergence. It does seem to be adding something valuable, as if you’ve discovered a magical ingredient by which you can explain mysterious phenomena. That’s the reason it continues to be popular, and gets inserted into scientific discussions. It convinces the listener that something has been explained with scientific rigour when all we’ve done is to say “it’s complicated”. > > Besides the good feeling, however, emergence is void of any explanatory power. And so it has no scientific value in a predictive capacity. You can’t use it to say anything about what an unknown system will do; only what you hope it will do. When applied to pie-in-the-sky AI futurism, emergence has become synonymous with “I’m sure the system will work itself out”. It indicates that the author has a feeling that a complex system will align at some point, but no clear sense of how, why, or when. Insofar as intelligence does manifest in a specific instance, “emergence” doesn’t tell us anything interesting about how it happened. And insofar as intelligence hasn’t yet manifested, emergence doesn’t tell us when it will or what direction to take to get there. > > In the field of AI development, emergence is invoked whenever someone encounters a phenomenon in the human mind and has no idea how to even start explaining it (e.g. art, socialization, empathy, transcendental aesthetics, DnD, etc). If said researcher already has a working theory of AI, this realization is disheartening. So they look deeper into the matter, find some point of overlap between the existing theory and the missing behaviour, and assume that with enough time and complexity the missing pieces will emerge. > > Emergence is attractive in such cases because it puts the author’s mind at ease, by making it seem like they have a viable mechanism that only needs more time to be vindicated. It placates their inner watchdog, the one that demands concrete, scientific explanations. Emergence, being related to complexity and superficially validated by experiments such as Conway’s Game of Life, is enough to lull that watchdog back to sleep. > > This justifies continuing to ignore any shortcomings in a theoretical model, and persisting on the current path. Like the proverbial man who searches for his lost keys under the lamplight, because that is where the light is, he hopes that with enough persistence his keys will “emerge”. The only other alternative is to admit failure, and to give up any hope of accomplishing what you want within this lifetime. > > Scientists, it seems, can have superstitions too. And emergence has a powerful narcotic effect: it feels so reasonable and credible on a gut level². There are many factors that prevent a given researcher from investigating emergence too deeply and realizing that it lacks any substance. First, there appears to be a lot of external evidence to back it up in the natural world. This, as was pointed out, equivocates between retrospective and prospective uses of the term, and so legitimate uses are being conscripted to justify the illegitimate ones. Secondly, the fact that emergence exclusively concerns itself with intractably complex systems means anything behind its curtain by definition can’t be studied. So it conveniently excludes itself from exactly that analysis which would reveal it to be hollow. > > In the end emergence isn’t an explanation; it’s an observation combined with a recognition of ignorance. Wherever emergence shows up there is an implicit acceptance that everyone involved is at a loss for how to approach the topic. It’s not that properties like intelligence won’t emerge from neural activity, it’s that emergence is a placeholder that justifies and promotes a lack of interest in exploring the details behind the connection. It discourages investigation. By invoking the term, we are merely thanking the nature gods for granting us this emergent property (aka property), and trying not to examine their gifts too profanely or with ingratitude. This impulse is understandable, since we don’t think we’ll discover an answer if we were to dig in. But we shouldn’t allow our insecurities to masquerade as science, or else they may become ingrained to the extent that they are difficult to uproot. A false answer stands in the way of a true one. > > ¹ This used to say ‘You can remove “emergent” from any sentence and it would mean the same thing’, but that has caused some confusion, so to clarify: the word “emergent” when used as an adjective doesn’t add new or useful information; you won’t know any more about the subject than you did before. > > ² A self-aware researcher should notice if they have a strong intuitive or emotional reason for holding on to the idea. If you ever feel that emergence is so self-evident that it can never be disproved, that should give you pause — perhaps you have strayed outside the bounds of scientific inquiry and into metaphysical expositions. Not that there’s anything wrong with the latter… > > > https://ykulbashian.medium.com/emergence-isnt-an-explanation-it-s-a-prayer-…

July 22, 2023

Prima discussione del Consiglio di Sicurezza ONU sull'AI

by Diego.Latella

Buona sera, Non mi pare sia passata sulla lista la notizia in ogetto. In caso contrario mi scuso per il doppione. In ogni caso, il video della riunione e' disponibile al seguente link: Artificial intelligence: opportunities and risks for international peace and security - Security Council, 9381st meeting | UN Web TV [1] Buona serata Diego -- Dott. Diego Latella - Senior Researcher CNR/ISTI, Via Moruzzi 1, 56124 Pisa, Italy (http:www.isti.cnr.it [2]) FM&&T Lab. (http://fmt.isti.cnr.it) CNR/GI-STS (http://gists.pi.cnr.it) https://www.isti.cnr.it/People/D.Latella - ph: +390506212982, fax: +390506212040 =================== The quest for a war-free world has a basic purpose: survival. But if in the process we learn how to achieve it by love rather than by fear, by kindness rather than compulsion; if in the process we learn how to combine the essential with the enjoyable, the expedient with the benevolent, the practical with the beautiful, this will be an extra incentive to embark on this great task. Above all, remember your humanity. -- Sir Joseph Rotblat I don't quite know whether it is especially computer science or its subdiscipline Artificial Intelligence that has such an enormous affection for euphemism. We speak so spectacularly and so readily of computer systems that understand, that see, decide, make judgments, and so on, without ourselves recognizing our own superficiality and immeasurable naivete with respect to these concepts. And, in the process of so speaking, we anesthetise our ability to evaluate the quality of our work and, what is more important, to identify and become conscious of its end use. […] One can't escape this state without asking, again and again: "What do I actually do? What is the final application and use of the products of my work?" and ultimately, "am I content or ashamed to have contributed to this use?" -- Prof. Joseph Weizenbaum ["Not without us", ACM SIGCAS 16(2-3) 2--7 - Aug. 1986] Links: ------ [1] https://media.un.org/en/asset/k1j/k1ji81po8p [2] http://www.isti.cnr.it

July 21, 2023

Junk websites filled with AI-generated text are pulling in money from programmatic ads

by Alberto Cammozzo

<https://www.technologyreview.com/2023/06/26/1075504/junk-websites-filled-wi…> More than 140 brands are advertising on low-quality content farm sites—and the problem is growing fast. People are using AI chatbots to fill junk websites with AI-generated text that attracts paying advertisers, according to a new report from the media research organization NewsGuard that was shared exclusively with MIT Technology Review. Over 140 major brands are paying for ads that end up on unreliable AI-written sites, likely without their knowledge. Ninety percent of the ads from major brands found on these AI-generated news sites were served by Google, though the company’s own policies prohibit sites from placing Google-served ads on pages that include “spammy automatically generated content.” The practice threatens to hasten the arrival of a glitchy, spammy internet that is overrun by AI-generated content, as well as wasting massive amounts of ad money. Most companies that advertise online automatically bid on spots to run those ads through a practice called “programmatic advertising.” Algorithms place ads on various websites according to complex calculations that optimize the number of eyeballs an ad might attract from the company’s target audience. As a result, big brands end up paying for ad placements on websites that they may have never heard of before, with little to no human oversight. To take advantage, content farms have sprung up where low-paid humans churn out low-quality content to attract ad revenue. These types of websites already have a name: “made for advertising” sites. They use tactics such as clickbait, autoplay videos, and pop-up ads to squeeze as much money as possible out of advertisers. In a recent survey, the Association of National Advertisers found that 21% of ad impressions in their sample went to made-for-advertising sites. The group estimated that around $13 billion is wasted globally on these sites each year. Now, generative AI offers a new way to automate the content farm process and spin up more junk sites with less effort, resulting in what NewsGuard calls “unreliable artificial intelligence–generated news websites.” One site flagged by NewsGuard produced more than 1,200 articles a day. Some of these new sites are more sophisticated and convincing than others, with AI-generated photos and bios of fake authors. And the problem is growing rapidly. NewsGuard, which evaluates the quality of websites across the internet, says it’s discovering around 25 new AI-generated sites each week. It’s found 217 of them in 13 languages since it started tracking the phenomenon in April. NewsGuard has a clever way to identify these junk AI-written websites. Because many of them are also created without human oversight, they are often riddled with error messages typical of generative AI systems. For example, one site called CountyLocalNews.com had messages like “Sorry, I cannot fulfill this prompt as it goes against ethical and moral principles … As an AI language model, it is my responsibility to provide factual and trustworthy information.” NewsGuard’s AI looks for these snippets of text on the websites, and then a human analyst reviews them. Making money from junk “It appears that programmatic advertising is the main revenue source for these AI-generated websites,” says Lorenzo Arvanitis, an analyst at NewGuard who has been tracking AI-generated web content. “We have identified hundreds of Fortune 500 companies and well-known, prominent brands that are advertising on these sites and that are unwittingly supporting it.” “This is not normal. This is not healthy.” MIT Technology Review looked at the list of almost 400 individual ads from over 140 major brands that NewsGuard identified on the AI-generated sites that served programmatic ads, which included companies from many different industries including finance, retail, auto, health care, and e-commerce. The average cost of a programmatic ad was $1.21 per thousand impressions as of January 2023, and brands often don’t review all the automatic placements of their advertisements, even though they cost money. Google’s programmatic ad product, called Google Ads, is the largest exchange and made $168 billion in advertising revenue last year. The company has come under criticism for serving ads on content farms in the past, even though its own policies prohibit sites from placing Google-served ads on pages with “spammy automatically generated content.” Around a quarter of the sites flagged by NewsGuard featured programmatic ads from major brands. Of the 393 ads from big brands found on AI-generated sites, 356 were served by Google. “We have strict policies that govern the type of content that can monetize on our platform,” Michael Aciman, a policy communications manager for Google, told MIT Technology Review in an email. “For example, we don’t allow ads to run alongside harmful content, spammy or low-value content, or content that’s been solely copied from other sites. When enforcing these policies, we focus on the quality of the content rather than how it was created, and we block or remove ads from serving if we detect violations.” Most ad exchanges and platforms already have policies against serving ads on content farms, yet they “do not appear to uniformly enforce these policies,” and “many of these ad exchanges continue to serve ads on [made-for-advertising] sites even if they appear to be in violation of … quality policies,” says Krzysztof Franaszek, founder of Adalytics, a digital forensics and ad verification company. Google said that the presence of AI-generated content on a page is not an inherent violation. “We also recognize that bad actors are always shifting their approach and may leverage technology, such as generative AI, to circumvent our policies and enforcement systems,” said Aciman. A new generation of misinformation sites NewsGuard says that most of the AI-generated sites are considered “low quality” but “do not spread misinformation.” But the economic dynamic of content farms already incentivizes the creation of clickbaity websites that are often riddled with junk and misinformation, and now that AIs can do the same thing on a bigger scale, it threatens to exacerbate the misinformation problem. For example, one AI-written site, MedicalOutline.com, had articles that spread harmful health misinformation with headlines like “Can lemon cure skin allergy?” “What are 5 natural remedies for ADHD?” and “How can you prevent cancer naturally?” According to NewsGuard, advertisements from nine major brands, including the bank Citigroup, the automaker Subaru, and the wellness company GNC, were placed on the site. Those ads were served via Google. Adalytics confirmed to MIT Technology Review that ads on Medical Outline appeared to be placed via Google as of June 24. We reached out to Medical Outline, Citigroup, Subaru, and GNC for comment over the weekend, but the brands have not yet replied. After MIT Technology Review flagged the ads on Medical Outline and other sites to Google, Aciman said Google had removed ads that were being served on many of the sites “due to pervasive policy violations.” The ads were still visible on Medical Outline as of June 25. “NewsGuard's findings shed light on the concerning relationship between Google, ad tech companies, and the emergence of a new generation of misinformation sites masquerading as news sites and content farms made possible by AI,” says Jack Brewster, the enterprise editor of NewsGuard. “The opaque nature of programmatic advertising has inadvertently turned major brands into unwitting supporters, unaware that their ad dollars indirectly fund these unreliable AI-generated sites." Related Story A high wire act looking down into their safety net. The nearest person has a inappropriate content flag in one hand. Catching bad content in the age of AI Why haven’t tech companies improved at content moderation? Franaszek says it’s still too early to tell how the AI-generated content will affect the programmatic advertising landscape. After all, in order for those sites to make money, they still need to attract humans to their content, and it’s currently not clear whether generative AI will make that easier. Some sites might draw in only a couple of thousand views each month, making just a few dollars. “The cost of content generation is likely less than 5% of the total cost of running a [made-for-advertising] site, and replacing low-cost foreign labor with an AI is unlikely to significantly change this situation,” says Franaszek. So far, there aren’t any easy solutions, especially given that advertising props up the entire economic model of the internet. “What is key to remember is that programmatic ads—and targeted ads more generally—are a fundamental enabler of the internet economy,” says Hodan Omaar, senior AI policy advisor at the Information Technology and Innovation Foundation, a think tank in Washington, DC. “If policymakers banned the use of these types of ad services, consumers would face a radically different internet: more ads that are less relevant, lower-quality online content and services, and more paywalls,” Omaar says. “Policy shouldn’t be focused on getting rid of programmatic ads altogether, but rather on how to ensure there are more robust mechanisms in place to catch the spread of misinformation, whether it be direct or indirect.”

July 21, 2023

Re: [nexa] LLM (di serie "A"): possibili in Italia/Europa?

by Damiano Verzulli

Il 21/07/23 10:07, Antonio ha scritto: > [...] > No, non è tanto, per niente tanto. > E' questo quello che viene fuori. > [...] > C'è qualcosa di sbagliato in questa analisi? Se sì, dove? ....stai a vedere che, piano piano... viene fuori che *anche* in ambito ricerca/LLM lo scenario è lo stesso di quello che sostengo essere per l'ambito "cloud". Ossia: per il "cloud".... tutti sostengono che gli hyperscaler sono inarrivabili e che --in Italia/Europa-- *NON* siamo in grado di allestire alternative.... mentre io sostengo che NON è esattamente cosi' [1]. Va a finire che.... con un DGX H100 di UniPI si possono gia' fare danni "enormi". Figuriamoci con una settimana di tempo/calcolo di Leonardo.... Chissa' che i problemi non siano tecnologici, economici o "metallici" (la quantita' di ferro necessaria....), ma siano altri. Felice di essere smentito... Bye, DV [1] https://server-nexa.polito.it/pipermail/nexa/2021-May/046942.html -- Damiano Verzulli e-mail: damiano(a)verzulli.it --- possible?ok:while(!possible){open_mindedness++} --- "...I realized that free software would not generate the kind of income that was needed. Maybe in USA or Europe, you may be able to get a well paying job as a free software developer, but not here [in Africa]..." -- Guido Sohne - 1973-2008 http://ole.kenic.or.ke/pipermail/skunkworks/2008-April/005989.html

July 21, 2023

Re: [nexa] LLM (di serie "A"): possibili in Italia/Europa? [Era: nexa Digest, Vol 171, Issue 53]

by Antonio

> Mah, spiegato così ha un senso, ma contiene l'errore tipico delle > estrapolazioni arbitrarie, quello di considerarle lineari, salvo poi > ridurre di un numero a piacere ed arbitrario di ordini di grandezza il > risultato, e concludere che è sempre tanto. No, non è tanto, per niente tanto. E' questo quello che viene fuori. Il 10^23 indicato dal prof. Attardi è di picco o totale? Nel primo caso, ho dimostrato che ci vorrebbero 15000 centrali elettriche per farlo girare, quindi non credo. Se è totale, allora bisogna conoscere il tempo impiegato nel calcolo. 1 giorno, 1 settimana, 1 mese? Il numero, sì questo, "arbitrario", che ho considerato è di due mesi. In questo caso il 10^23 si riduce ad un "misero" 19 petaFLOPS di picco. (i due NVIDIA DGX H100 acquistati da UniPI) hanno ciascuno performance di 32 petaFLOPS. C'è qualcosa di sbagliato in questa analisi? Se sì, dove? Grazie, Antonio

July 21, 2023

Re: [nexa] LLM (di serie "A"): possibili in Italia/Europa? [Era: nexa Digest, Vol 171, Issue 53]

by Marco A. Calamari

On gio, 2023-07-20 at 10:19 +0200, Damiano Verzulli wrote: > > Il 20/07/23 00:39, Marco A. Calamari ha scritto: > > On mer, 2023-07-19 at 18:26 +0200, Antonio wrote: > > > Più per capire io ... > > > Prendiamo per buono 10^23 (come fonte "secondaria" ho trovato questa [1]) > > > 10^23 FLOPS, convertiti, sono 100.000 exaFLOPS. > > > Frontier, il primo supercomputer della lista dei top100, supera di poco 1 > > > exaFLOPS e > > > richiede 22 MW di potenza. > > > 22 MW * 100000 = 2200000 MW = 22000 GW ovvero 15000 centrali elettriche > > > (1520 MW è la potenza della più grande centrale elettrica italiana [2]) > > > messe > > > assieme. > > > Se non ho cannato i calcoli, mi sembra un poco eccessivo. > > Li hai cannati di brutto. > > 22 MW è la potenza di una molto piccola centrale elettrica. > > Tipo la vecchia diga sul fiume Lima qui in toscana. > > O la potenza installata in 8000 appartamenti > > Negativo. Ho impiegato un po' a fare il reverse-engineering della tesi > di Antonio... ma è corretta ( i calcoli, sono corretti). > > Il ragionamento fatto da lui e': > > a - la potenza di calcolo di riferimento indicata da Attardi è 10^23 > FLOPS, che è circa 100.000 volte superiore a quella indicata dal > supercomputer al numero 1 della top500, che dichiara consumi per 22MW; > > b - spannometricamente, quindi, si assume che se serve un sistema HPC > 100.000 volte piu' potente di uno che consuma 22MW, si puo' supporre che > questo consumera' 100.000 volte i consumi di quello da 22MW; > > c - 100.000 volte 22MW fanno 22.000 GW > > d - si puo' supporre, quindi, che il sistema HPC da cui siamo partiti > (quello da 10^23) consuma 22.000 GW, ossia l'equivalente di energia > prodotta da 15.000 centrali elettriche. > > Personalmente, trovo il ragionamento "condivisibile", posto che > esisteranno N "margini per economie [energetiche] di scal". Ma se anche > riduciamo di un fattore 100 i consuni... otteniamo sempre 150 centrali.... > > Temo ci sia qualcosa che sfugge ai nostri ragionamenti (Antonio/mio), > oppure ci sono "errori" in 10^23... Mah, spiegato così ha un senso, ma contiene l'errore tipico delle estrapolazioni arbitrarie, quello di considerarle lineari, salvo poi ridurre di un numero a piacere ed arbitrario di ordini di grandezza il risultato, e concludere che è sempre tanto. Ne ho viste molte con questa "logica" ... IMHO, of course. > _______________________________________________ > nexa mailing list > nexa(a)server-nexa.polito.it > https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa

July 20, 2023

Re: [nexa] Authors call for AI companies to stop using their work without consent | Books | The Guardian

by Stefano Maffulli

On Thu, Jul 20, 2023 at 11:26 AM Fabio Alemagna <falemagn(a)gmail.com> wrote: > Non vedo ragione per la quale a questo giro dovrebbe andare diversamente. > > Creative Commons, che di copyright e etica annessa se ne intende, la > vede alla stessa maniera: > https://creativecommons.org/2023/02/17/fair-use-training-generative-ai/ Purtroppo la conversazione pubblica si sta incancrenendo con i poveri autori (e la società in generale direi) da un lato, quelli che hanno prodotto e i riccastri BigTech che hanno accumulato di tutto dall'altro lato. La simpatia per BigTech non ce l'ha nessuno, giustamente. E si sparge l'idea che BigTech ruba, e ognuno vuole un lucchetto. Nel marasma delle urla di autori che vogliono espandere ancora di più il copyright però si perde di vista che il copyright si è esteso da decenni e non sono stati certo gli autori a guadagnarci. Per me sarebbe meglio indebolire il copyright per aumentare la possibilità di accumulare dati per generare AI da condividere nel commons. Il lavoro che fanno gruppi non-profit come Eleuther AI e LAION potrebbero portare ad avere LibreOffice, Krita, GIMP e compagnia ad avere funzioni avanzate invece di rimanere ancora più irrimediabilmente indietro.

July 20, 2023

Re: [nexa] LLM (di serie "A"): possibili in Italia/Europa? [Era: nexa Digest, Vol 171, Issue 53]

by Antonio

> Temo ci sia qualcosa che sfugge ai nostri ragionamenti (Antonio/mio), > oppure ci sono "errori" in 10^23... Credo di aver capito dove sbaglio / sbagliamo. 10^23 sono FLOPS *totali*, per l'intero periodo di calcolo. Ad esempio, se la "macchina" ha girato per 2 mesi: 60 (secondi) * 60 (minuti) * 24 (ore) * 60 (giorni) = 5184000 secondi 10^23 / 5184000 = 1,9 * 10^16, sempre un numero enorme, ma non 15000 centrali elettriche ;) Antonio

July 20, 2023