Intelligence artificielle & cie
Au cours de l'été 2024, j'ai commencé une veille [critique] à propos de l'intelligence artificielle et tout autre machin connexe.
Pour chaque hyperlien, j'extrais une citation qui résume bien le sujet ou une partie du texte qui m'a marqué.
2024-12-03
The conversation no one wants to have is that it’s still impossible to calculate how many resources the infrastructure of AI needs to run at scale—including water, a resource Spain is rapidly running out of, with 78% of its landmass threatened by desertification. By choosing AI industry-driven growth today, will urban planners in Spain jeopardize their water tomorrow?
[…]
A Washington Post investigation conducted in collaboration with researchers at the University of California, Riverside, recently revealed that ChatGPT requires up to three bottles of water to generate a single 100-word email.
— https://www.techpolicy.press/why-we-dont-know-ais-true-water-footprint/
2024-12-02
«Le volume de données extraites est désormais tel que des résultats relevant de la science-fiction semblent aujourd’hui possibles. Mais parce qu’il s’agit d’une industrie extractive, et parce qu’elle doit fonctionner à une échelle aussi vaste, il y a des limites strictes à ce que les technologies actuelles de l’IA peuvent faire. C’est ainsi que la valorisation boursière des entreprises technologiques dépasse largement la réalité économique, créant les conditions d’une bulle spéculative classique.
Le premier obstacle est l’épuisement de la matière première que constituent les données humaines. Selon un calcul publié dans le Wall Street Journal, l’IA sera à court de données provenant de l’ensemble de l’internet et produites par l’ensemble de l’humanité dès 2026. Les entreprises spécialisées dans l’IA ont donc pris l’habitude d’utiliser des données générées par l’IA pour essayer d’entraîner leurs machines ! Mais cela produit ce qu’un récent article universitaire a appelé « l’effondrement du modèle » : L’IA cesse de fonctionner lorsqu’elle doit se nourrir d’elle-même. Et plus l’internet sera inondé de « déchets d’IA », moins l’IA sera utile.
[…]
Les centres de données se multiplient dans le monde entier pour répondre à la demande : Microsoft ouvre actuellement un nouveau datacenter quelque part sur la planète tous les trois jours.
— https://lvsl.fr/pourquoi-la-bulle-de-lintelligence-artificielle-est-condamnee-a-exploser/
2024-10-08
«Researchers — including many women of color — have been saying for years that these systems interact differently with people of color and that the societal effects could be disastrous: that they’re a fun-house-style distorted mirror magnifying biases and stripping out the context from which their information comes; that they’re tested on those without the choice to opt out; and will wipe out the jobs of some marginalized communities.
Gebru and her colleagues have also expressed concern about the exploitation of heavily surveilled and low-wage workers helping support AI systems; content moderators and data annotators are often from poor and underserved communities, like refugees and incarcerated people. Content moderators in Kenya have reported experiencing severe trauma, anxiety, and depression from watching videos of child sexual abuse, murders, rapes, and suicide in order to train ChatGPT on what is explicit content. Some of them take home as little as $1.32 an hour to do so.
[…]
In other words, the problems with AI aren’t hypothetical. They don’t just exist in some SkyNet-controlled, Matrix version of the future. The problems with it are already here.»
«Former Google CEO Eric Schmidt says we should go all in on building AI data centers because ‘we are never going to meet our climate goals anyway'»
— https://www.businessinsider.com/eric-schmidt-google-ai-data-centers-energy-climate-goals-2024-10
2024-10-03
«Microsoft claims its new tool can correct AI hallucinations, but experts advise caution
“Trying to eliminate hallucinations from generative AI is like trying to eliminate hydrogen from water,” said Os Keyes, a PhD candidate at the University of Washington who studies the ethical impact of emerging tech. “It’s an essential component of how the technology works.”»
2024-10-01
«Les êtres humains cachés derrière l'IA en Amérique latine
[…]
“Dans le capitalisme, tout est sous-traité”, ajoute Álvaro Montes. Ce n'est pas une exception. Dans de nombreux cas, l'étiquetage des données est effectué par des sous-traitants de la Silicon Valley, qui à leur tour sous-traitent ces tâches à d'autres entreprises. Cette externalisation rend plus difficiles la syndicalisation et la contestation des pratiques contraires à l'éthique, ainsi que la revendication de meilleures conditions de travail.»
— https://fr.globalvoices.org/2024/09/30/290760/
2024-09-26
«OpenAI asked US to approve energy-guzzling 5GW data centers, report says
[…]
The AI company’s CEO, Sam Altman, supposedly pitched the plan after a recent meeting with the Biden administration where stakeholders discussed AI infrastructure needs. Bloomberg reviewed an OpenAI document outlining the plan, reporting that 5 gigawatts “is roughly the equivalent of five nuclear reactors” and warning that each data center will likely require “more energy than is used to power an entire city or about 3 million homes."»
2024-09-23
In Southwest Memphis, Musk isn’t using a single methane generator. He’s using at least 18
As Dara Kerr reported for NPR on September 11, Musk has taken over a former Electrolux manufacturing facility on 217 acres of land, with the possibility of taking over an extra 580 acres. His plan is to build the “world’s largest supercomputer.” He’s calling it “Colossus.” It will power Musk’s ambitious AI-startup, xAI.
— https://unprecedented.ghost.io/archive/poisoning-southwest-memphis/
2024-09-18
«A bottle of water per email: the hidden environmental costs of using AI chatbots»
— https://www.washingtonpost.com/technology/2024/09/18/energy-ai-use-electricity-water-data-centers/
«Ellison declares Oracle all-in on AI mass surveillance, says it'll keep everyone in line
Cops to citizens will be ‘on their best behavior because we're constantly recording and reporting»
— https://www.theregister.com/2024/09/16/oracle_ai_mass_surveillance_cloud/
2024-09-17
«The Minecraft simulation also demonstrated how AI agents can change their minds at the drop of the hat—as the successful cult bribing incident suggests. In other instances, a farmer needed for the food supply decided spontaneously to give up agriculture and go on an adventure. In another instance, an entire village stopped working because of a single missing bot.
Right now this is happening in test environments. But soon AI agents will be changing the real world—and at a pace none of us are prepared for.»
— https://www.honest-broker.com/p/eleven-predictions-heres-what-ai
2024-09-16
«On observe une adéquation entre le développement de l’IA et les priorités des géants pharmaceutiques. Les ressources sont orientées vers la production de technologies hyperspécialisées en oncologie ou en génétique. Pourtant, la recherche montre que des politiques orientées vers des investissements massifs dans des traitements curatifs sont inefficaces du point de vue de la santé publique. Des actions préventives axées sur l’environnement ou le logement le sont significativement plus. C’est une vision de l’efficacité que ceux qui sont à l’origine des initiatives en IA en santé ne partagent pas.»
2024-09-15
«I call it algorithmic management. It is a simple idea with big implications, made possible by the vast computational power now available to rich companies. Algorithmic management refers to a way of organising work in which workers are directed, monitored, tracked and assessed – all at once, in real time – by a computer system that doesn’t rely on a human manager to control it. Far beyond its application in headline-grabbing companies such as Amazon, it is a way of organising work with wide-ranging consequences.»
— https://www.opendemocracy.net/en/ai-tech-algorithms-work-class-amazon-uber/
2024-09-14
«The largest of these models are incredibly expensive. They’re expensive to make, with billions spent acquiring training data, labelling it, and running it through massive computing arrays to turn it into models.
Even more important, these models are expensive to run. Even if a bankrupt AI company’s model and servers could be acquired for pennies on the dollar, even if the new owners could be shorn of any overhanging legal liability from looming copyright cases, even if the eye-watering salaries commanded by AI engineers collapsed, the electricity bill for each query – to power the servers and their chillers – would still make running these giant models very expensive.»
— https://locusmag.com/2023/12/commentary-cory-doctorow-what-kind-of-bubble-is-ai/
2024-09-12
«Google’s NotebookLM app can now generate ‘lively’ audio discussions with two AI hosts about the documents you’ve given it.»
— https://www.theverge.com/2024/9/11/24242138/google-notebook-llm-ai-fake-podcasts-research
2024-09-11
«Why has the Internet of Things failed? According to a survey last year, less than 50% of appliances that are internet-capable ever get connected. When I talk to manufacturers, I often hear even worse numbers, sometimes below 30%! Despite many years and billions of dollars of investment into the “Internet of Things”, this lack of adoption makes it clear that even if a device can be connected, consumers don’t see the value in most cases. I think it’s time to admit that the core idea of IoT has failed. To understand why, it’s worth looking at how it was originally pitched, and what flaws time has revealed in those arguments.»
— https://petewarden.com/2024/08/23/why-has-the-internet-of-things-failed/
2024-09-09
«Shady Firms Say They're Already Manipulating Chatbots to Say Nice Things About Their Clients
[…]
The goal? Manipulate AI chatbots and search tools into holding favorable views of their clients, which might be brands, websites, or humans. The idea is that when someone uses an AI tool to look for info about these clients, the AI model can be tweaked into offering a glowing review.»
— https://futurism.com/shady-firms-manipulating-chatbots
2024-09-03
«Canva says its AI features are worth the 300 percent price increase»
— https://www.theverge.com/2024/9/3/24234698/canva-price-increase-300-percent-ai-features
2024-08-29
«According to experts, OpenAI’s business model is “untenable”, the company is on course to lose nearly $5B by the end of the year—it makes an annualized revenue of $3.4B, but has spent $8.5B on AI training and staff—and it will not make it to 2026 if it doesn’t raise more capital than any other start-up has ever raised.»
— https://www.aitoolreport.com/articles/openais-desperate-funding-race
2024-08-24
«Alors que les signes de l'éclatement de la bulle financière de l'IA s'amoncellent, le marché continue de dépenser comme jamais. Le signe d'un régime politique et économique délirant, dans lequel une poignée de monopoles s'exonère de la rationalité.»
— https://www.arretsurimages.net/chroniques/clic-gauche/monopoly
2024-08-23
«Copilot AI calls journalist a child abuser, Microsoft tries to launder responsibility»
2024-08-22
«Only a few short months ago, Generative AI was sold to us as inevitable by the leadership of AI companies, those who partnered with them, and venture capitalists. As certain elements of the media promoted and amplified these claims, public discourse online buzzed with what each new beta release could be made to do with a few simple prompts. As AI became a viral sensation, every business tried to become an AI business. Some businesses added “AI” to their names to juice their stock prices, and companies talking about “AI” on their earnings calls saw similar increases. While the Generative AI hype bubble is now slowly deflating, its harmful effects will last.»
— https://arxiv.org/abs/2408.08778
2024-08-07
«While tech giants have been adamant that AI is a long game, analysts are becoming increasingly wary of the losses.
Last week, The Information reported that OpenAI could end up losing $5 billion this year alone, and could run out of cash in the next 12 months without any major cash injections.
“Despite its expensive price tag, the technology is nowhere near where it needs to be in order to be useful,” Goldman Sachs' most senior stock analyst Jim Covello argued in a report last month.
Google has similarly had to reassure investors, with CEO Sundar Pichai saying that the company was “at the early stage of what I view as a very transformative area” during the company’s earnings call last week.
Much like Microsoft, Google has rapidly ramped up spending on AI infrastructure in the first half of this year, and is expected to spend $49 billion in capital expenditures by the end of the year.»
— https://futurism.com/the-byte/microsoft-losing-money-ai
2024-08-02
«It can be easy to get lost in numbers, but what they point to is an obvious truth: generative AI is an environmental disaster that’s accelerating natural destruction and the climate crisis at the very moment alarms are sounding about the precious little time that remains to turn things around. Tech companies once pitched themselves as the purveyors of a more ethical form of capitalism. They wanted us to believe they would balance corporate profit with environmental sustainability, such that the digital future was marketed as inherently more sustainable than the analog past. It’s clearer than ever that was a lie.»
— https://disconnect.blog/generative-ai-is-a-climate-disaster/
2024-07-31
«From a labor perspective, AI is really an “extraction machine.” Beneath the polished exterior of our devices lies the complex network of components and relationships necessary to power it. When AI breaks down or does not function properly, human workers are there to step in and assist algorithms in completing the work. When Siri does not recognize a voice command or when facial recognition software fails to verify a person’s identity, these cases are often sent to human workers to establish what went wrong and how the algorithm could be improved.»
— https://www.siliconvalley.com/2024/07/30/opinion-whats-behind-the-ai-boom-exploited-humans/amp/
2024-07-30
Microsoft’s co-founder, Bill Gates, said last week that AI would help combat the climate crisis because big tech is “seriously willing” to pay extra to use clean electricity sources in order “to say that they're using green energy”.
[…]
Water usage is another environmental factor in the AI boom, with one study estimating that AI could account for up to 6.6bn cubic metres of water use by 2027 – nearly two-thirds of England’s annual consumption.»
— https://www.theguardian.com/technology/article/2024/jul/02/google-ai-emissions
2024-07-29
«The ClaudeBot web crawler that Anthropic uses to scrape training data for AI models like Claude has hammered iFixit’s website almost a million times in a 24-hour period, seemingly violating the repair company’s Terms of Use in the process.»
2024-07-19
«Proton Mail is a privacy-focused email service. It’s the level of privacy service that privacy obsessives recommend to their friends.
Proton Mail ran a user survey two months ago. They found some readers saying they were “interested in AI,” didn’t include a “hell no” option, and today, they’ve introduced Proton Scribe, claiming that “interested in AI” constituted user demand for this specific feature!»
— https://pivot-to-ai.com/2024/07/18/proton-mail-goes-ai-security-focused-userbase-goes-what-on-earth/
2024-07-09
«OpenAI Pleads That It Can’t Make Money Without Using Copyrighted Materials for Free “It would be impossible to train today’s leading AI models without using copyrighted materials."»
— https://futurism.com/the-byte/openai-copyrighted-material-parliament, Noor Al-Sibai
2024-07-08
«The tech giant, which has seen its planet-warming emissions rise because of artificial intelligence, has stopped buying cheap offsets behind the neutrality claim. The company now aims to reach net-zero carbon by 2030.»
— https://www.bloomberg.com/news/articles/2024-07-08/google-is-no-longer-claiming-to-be-carbon-neutral, Akshat Rathi
2024-07-02
«Meta a annoncé qu’à partir du 26 juin, Facebook et Instagram allaient se servir de tout ce qui est sur leurs plateformes pour entraîner leur AI. Ça veut dire mes dessins, vos dessins, vos photos de 2014 et celles de vos chiens, vos chats et oups, de vos enfants aussi!»
— https://cabtastic.substack.com/p/les-cabtualites-de-juin-59f, Cab
2024-06-29
«Partager le contenu de vos clients avec Figma pour la formation des modèles d'IA est facultatif. À partir du 15 août 2024, les administrateurs pourront contrôler le partage de contenu avec les clients grâce à un nouveau paramètre. Dans la mesure où aucune formation au contenu n'aura lieu avant cette date, vous avez le temps de prendre une décision. Le contenu client comprend les documents créés ou téléchargés dans Figma.»