🚀 Gigantic floating city design ; a first: IA fait jeu égal avec nous à Diplomacy, un jeu supposant négo et coopération, & plus !
🚀 Gigantic floating city design ; a first: IA fait jeu égal avec nous à Diplomacy, un jeu supposant négo et coopération, & plus !
AI-generated song got 100 millions streams ; s'habiller avec des écrans ; AI reunites Holocaust survivor with childhood photos
Bonjour,
Vous recevez la newsletter Parlons Futur : une fois par semaine au plus, une sélection de news résumées en bullet points sur des sujets tech 🤖, science 🔬, éco 💰, géopolitique 🌏 et défense ⚔️ pour mieux appréhender le futur 🔮.
Je m'appelle Thomas, plus d'infos sur moi en bas d'email.
Voici donc ma dernière sélection !
L’apéro
Exemple probant d'utilisation de GPT-3 à la place de Google "GPT-3 has basically replaced Google for me for searching through humanity's collective memory." (source)
attention toutefois, ces "language models" ne savent pas discerner le vrai du faux, résultats à manier avec prudence, voir le contre-exemple ci-dessous
AI not ready for prime time : Meta launched Galactica, a large language model for science. "It can summarize academic literature, solve math problems, generate Wiki articles, write scientific code, annotate molecules and proteins, and more." But it quickly backfired as it could be made to generate nonsense (examples), they even had to remove it after 3 days.
A fundamental problem with Galactica is that it is not able to distinguish truth from falsehood, a basic requirement for a language model designed to generate scientific text. People found that it made up fake papers (sometimes attributing them to real authors), and generated wiki articles about the history of bears in space as readily as ones about protein complexes and the speed of light. It’s easy to spot fiction when it involves space bears, but harder with a subject users may not know much about.
Critics said "Language models are not really knowledgeable beyond their ability to capture patterns of strings of words and spit them out in a probabilistic manner, it gives a false sense of intelligence.”
Yann LeCun, a Turing Award winner and Meta’s chief scientist, defended Galactica to the end. On the day the model was released, LeCun tweeted: “Type a text and Galactica will generate a paper with relevant references, formulas, and everything.” Three days later, he tweeted: “Galactica demo is off line for now. It’s no longer possible to have some fun by casually misusing it. Happy?”
Heureusement Meta / Yann LeCun ont une autre annonce cette semaine : Big AI milestone today: CICERO, an AI agent that can negotiate and cooperates with people
It is the first AI system to achieve human-level performance in the popular strategy game Diplomacy. (voir après l’apéro pour plus de détails sur ce sujet fascinant)
Tencent has a new AI music-generation project that released 1k songs in China - one got 100 millions streams. (source)
If streamed exclusively on Spotify, 100,000,000 streams would translate to around $350,000 in revenue, according to StreamingCalculator.
That AI tool can generate quizzes and assessments at scale from any text in one click using AI, for teachers, schools, HR teams: questgen.ai (80-sec demo video)
Exemples d’adaptation d’Unilever au changement climatique (The Economist) :
Unilever has laid out plans to relocate manufacturing if particular plants are damaged by extreme weather and has lined up emergency suppliers if supply-chains are disrupted.
It is developing longer-term contingency plans, too, such as making its shampoos quicker to rinse, in case its customers are obliged to curb their use of water.
Most of its dishwasher detergent works in cold water, in anticipation of a world where energy is much pricier.
Le potentiel de la géothermie : if we develop the capability to drill cost-effectively to depths of 10-20km in temperatures of 500°C, we can expand the number of sites suitable for geothermal power by 100 globally.
1000 gigawatts of geothermal power (roughly equivalent to all the solar panel capacity that has been installed globally to date, also equivalent to roughly 1000 nuclear reactors) would require about 30,000 wells (puits).
That is about half the number of wells the oil industry drills every year (albeit to shallower depths and with different technology). (tech investor Vinod Khosla)
Solar parking: France will require all parking lots for more than 80 cars to be covered by solar panels; the scheme is forecast to produce as much power as 11 gigawatts, equivalent to 10 nuclear reactors. (source)
Toujours aussi flippantes ces vidéos orwelliennes : vidéo de 45 secondes d'un drone avec haut-parleur survolant un quartier en Chine et intimant l'ordre aux habitants d'aller se faire tester sans délai
Filmmakers and advertising executives use DALL-E 2 for client pitches. (Peter Diamandis)
The New York Times recently interviewed several creative professionals on how they partner with AI to generate more creative ideas—faster. For instance, an Australia-based filmmaker, who has worked on hit shows including “Westworld,” is now using AI-generated art in his pitches to film studios.
In one example, he was looking for an image of a specific type of marble statue. But after an unsuccessful search on Getty Images, he turned to DALL-E 2 and the AI generated better ideas than what he could find on Getty in a fraction of the time.
As the filmmaker put it, “It’s like working with a really willful concept artist.”
LG’s new thin and stretchable displays could be used to wrap skin, clothing, cars, and furniture (source)
can be deformed by up to 20% of its original size and shape without suffering any damage.
a resolution that competes with most existing monitors
How fast can you speak English, without making mistakes? Take the Rap God Test (read a Wikipedia passage quickly) and get judged by an AI scoring your speech's intelligibility. Each error you make adds 1s to your time.
A new social media app for high schoolers has dethroned TikTok and BeReal in the App Store rankings — and is surprisingly not toxic (Business Insider)
The app is names "Gas", questions are asked, they're only positive, like "who should always DJ parties" and users can only respond sharing classmates' names
Super démo de 30 secondes : via un casque de réalité augmentée, expliquez avec vos mots à un assistant virtuel le motif de papier peint qui vous ferait plaisir, l'IA le génère et tapisse votre mur avec
faites vos remarques, l'IA revoit a copie, et passez commande dans la foulée :
"maïs", "non maïs en épi", "au format stylisé et parsemé de fleurs roses", "en plus grand", "cool, let's buy it"
Incroyable démo de 40 secondes : capturez en photo un objet autour de vous, l'IA le détoure, et collez-le dans la foulée sur un document ou un environnement de travail sur votre ordi
Nouvelle appli pour apprendre l'anglais grâce à l'IA "Speak out loud 10x more per lesson than any other method, Receive instant feedback from our AI tutor" (speak.com)
Ils ciblent d'abord les marchés coréen et japonais
“Under the hood, we combine the latest from OpenAI with in-house models to deliver the best performance across speech recognition, speech generation and conversation generation,”
Intelligence animale :
Julie, a chimpanzee studied in Zambia, stuck a piece of long grass in her ear, adjusted and wore it. The behaviour (a fashion trend, perhaps?) was soon sported by eight of the 12 members of the group. (The Economist)
Researchers have discovered that one population of white-faced capuchin monkeys in Panama are 3,000 years into their own 'Stone Age'. The monkeys have started using stone tools to break nuts and shellfish, making them the fourth type of primates to do so after us (source)
À votre bon coeur ❤️
Si vous appréciez cette synthèse gratuite, n’hésitez pas à prendre 3 secondes svp pour l’envoyer à ne serait-ce qu’un contact 🙂
Et si on vous a partagé cet email, vous pouvez cliquer ici pour vous inscrire et ne pas manquer les prochains
À table !
AI reunites Holocaust survivor with childhood photos (BBC)
Blanche Fixler remembers hiding inside a bed while Nazis searched for her.
"I felt them tapping on the bed," she recalls. "I said, you better not breathe or sneeze or anything - or you'll be dead."
Her mother and her siblings were murdered - but she was saved, thanks to her Aunt Rose, who hid her.
Now a tool using artificial intelligence (AI) - built by Daniel Patt, a software engineer for Google - could hold the key to putting names to some of the many faces, both victims and survivors, in hundreds of thousands of historic photographs. It found Blanche in a wartime photo which she had never seen before.
Daniel's website, Numbers to Names, uses facial recognition technology to analyse a person's face. It then searches through archive photos to find potential matches.
That detective work - joining the dots - could then help identify some people in photos whose identities are currently unknown.
Blanche immediately recognised herself standing at the front of the large group of people, but that's not all.
She also identified her Aunt Rose and one of the boys in the photo - giving Daniel and the United States Holocaust Memorial Museum new information to work with.
Before Blanche set eyes on the photo, only three people in it had been identified.
Thanks to Blanche, and Daniel's software, that number has doubled.
CICERO by Meta, the first AI system to achieve human-level performance in the popular strategy game Diplomacy (Meta, Science.org, The Economist)
Diplomacy is a board game where seven players compete to control supply centers (SCs) on a map, by moving their units into them. A player wins by controlling a majority of SCs.
CICERO demonstrated this by playing anonymously on webDiplomacy.net, an online version of the game, where CICERO achieved more than double the average score of the human players and ranked in the top 10% of participants who played more than one game.
Diplomacy, a strategy game which requires building trust, negotiating and cooperating with multiple players.
Diplomacy has been viewed for decades as a near-impossible grand challenge in AI because it requires players to master the art of understanding other people’s motivations and perspectives; make complex plans and adjust strategies; and then use natural language to reach agreements with other people, convince them to form partnerships and alliances, and more.
Unlike games like Chess and Go, Diplomacy is a game about people rather than pieces. If an agent can't recognize that someone is likely bluffing or that another player would see a certain move as aggressive, it will quickly lose the game. Likewise, if it doesn't talk like a real person -- showing empathy, building relationships, and speaking knowledgeably about the game -- it won't find other players willing to work with it.
Each turn's actions occur simultaneously after non-binding, private negotiations. To succeed, an agent must account for the risk that players may not stay true to their word, or that other players may themselves doubt the honesty of the agent. For this reason, an ability to reason about the beliefs, goals, and intentions of others and an ability to persuade and build relationships through dialogue are powerful skills in Diplomacy.
CICERO is so effective at using natural language to negotiate with people in Diplomacy that they often favored working with CICERO over other human participants.
CICERO has 2 components: a planning engine and a dialogue component
Un prof de Stanford ironise sur Twitter “Nothing to see here. Just an AI system learning to strategically plan world conquest, negotiate with humans in English and then betray them”
Trahir ? Pas encore en fait : Cicero can negotiate, convince, co-operate and compete. Seasoned Diplomacy players will, though, want to know something else: has it learned how to betray? But, though Cicero did, “strategically withhold information from players in gameplay”, it did not actually betray any of its opponents. Perhaps it was this final lack of Machiavellian ruthlessness which explains why it was only in the top 10%.
Gigantic floating city design (CNN, 3-min video on Youtube)
can house 60,000 guests
able to cruise at a speed of close to 10km/h
$8 billion to build, according to designers
550 meters long, and 610 meters wide
will have space for hotels, shopping malls, parks and even ports for smaller ships and aircraft
It'd take a special place to build it, too -- one that doesn't currently exist. The designers have posited Saudi Arabia as a location. Around one square kilometer of sea would need to be dredged and a circular dam constructed before the build could commence. The designers have earmarked a space at King Abdullah Port, 81 miles north of Jeddah as the ideal location.
la partie immergée serait haute de 30 mètres (draft of 30 meters)
Designers hope that construction could start in 2033, with a build time of eight years.
Cette plaque de cuisson électrique chauffe l'eau 10 fois plus vite qu'une plaque au gaz et peut alimenter votre maison en électricité (source)
bout un litre d'eau en 40 secondes vs 400 secondes, cf la démo
grâce à une batterie qui permet d'utiliser plus d'énergie électrique par seconde que ce qu'offre un branchement classique au secteur
It’s not easy to just replace gas and heating oil with electric power, since that requires doing expensive remodeling of the buildings that currently use those fuels. The wiring in many U.S. homes and businesses simply isn’t robust enough to handle the kind of power demands that a fully electrified building would require — you have to tear out the wiring and install better wiring, and that is difficult and expensive. Anyone who has installed an induction stove knows this all too well
But what if instead of rewiring a whole building, you could just buy battery-powered appliances and plug them into your standard outlet? That’s the basic idea behind Impulse. You put a battery in the appliance — the stove, or dryer, or oven, or heat pump, or whatever — and it can draw power slowly throughout the day through your regular 120V power outlet. And then when it comes time to cook your food, or dry your clothes, or cool down your house, or whatever, the appliance’s battery will release the stored electricity all in a rush. FWOOM! No fancy new wiring or massive investment program required.
If you happen to have a 240V outlet, it gets even fancier — the stove can actually supply power to the rest of the house.
This will make a large portion of building decarbonization cheap and easy
High power seems like it will also be a selling point for other electric appliances — dryers that can dry your clothes quickly, heating and air conditioning systems that can change the temperature of your house rapidly, and so on. This isn’t directly relevant to the climate, but it's likely that the advantages of battery-powered appliances over their traditional gas-powered cousins will speed up adoption.
à relativiser toutefois, "cooking represents only 2.8% of residential natural gas consumption in 2015 in the US (space heating 67%) (source)
but it seems that people demand to keep their gas lines because of the stove (les plaques de cuisson), and that means they incidentally use gas for all the other stuff, so it's great if we have a way to incentivize them to do without gas stoves.
AI Drew This Stunning Comic Series. You'd Never Know It (source)
You might expect a comic book series featuring art generated entirely by artificial intelligence to be full of surreal images that have you tilting your head trying to grasp what kind of sense-shifting madness you're looking at.
Not so with the images in The Bestiary Chronicles, a free, 3-part comics series
The visuals in the trilogy -- believed to be the first comics series made with AI-assisted art -- are stunning. They're also stunningly precise, as if they've come straight from the hand of a seasoned digital artist with a very specific story and style in mind.
"We're seeing the rise of a completely new visualization tool that will radically change the storytelling process across both the comics industry and entertainment in general," said Steve Coulson, writer of the trilogy
For The Bestiary Chronicles, Coulson turned to Midjourney, a service that quickly turns short text phrases, or "prompts," into images by scanning a giant database trained on visual art by humans.
The Bestiary Chronicles is a 114-page science fiction odyssey about monsters born from man's technological hubris. But it also showcases the remarkable progress of products like Midjourney, which are producing increasingly more sophisticated and refined images.
Coulson doesn't think tools like Midjourney will replace the comics artists he's long loved. "Those geniuses have an eye for dramatic composition and dynamic narrative that I strongly doubt machine learning will be able to match," he writes in the afterword to Summer Island. "But as a visualization tool for nonartists like myself, it's a hell of a lot of fun."
He does, however, see Midjourney as his true collaborator in The Bestiary Chronicles, even giving it an author credit.
Where a comics artist might conceive of a narrative and then create art to illustrate it, AI-assisted images have the potential to more actively steer the story, or even change its direction, thus dramatically redefining the whole creative workflow. Coulson likens this human-machine duet to improv jazz.
"I would never ask a human artist to just 'draw 100 splash pages and maybe I'll pick the one I like the best,' but Midjourney will happily spit them out 24/7," Coulson says. "Then after we review the imagery, we start to assemble the story, almost as an act of collage, filling in gaps along the way."
AI art is the star here, but humans had the decisive hand in which images made it into the final version of each story. They experimented with text prompts and carefully selected their final images from multiple Midjourney offerings, making a Photoshop tweak here and there, but mostly letting the machine-made work stand.
"We also used the phrase 'Hitchcock Blonde' to describe our heroine, and more often than not she came out looking like Grace Kelly," Coulson said. That's a fully recognizable Grace Kelly, without misplaced ears or a dog snout.
"The advances in AI image generation over the last few months have been exponential and mind blowing," Coulson said, "and this technology is only going to get better -- faster than we can imagine."
Si tout échoue, la dernière solution pour contrer le réchauffement climatique : pulvériser des aérosols dans la haute atmosphère pour renvoyer vers l'espace une partie du rayonnement solaire (Wired)
aérosol, définition : dispersion en particules très fines d'un liquide, d'une solution ou d'un solide dans un gaz.
Stratospheric Aerosol Injection (SAI) would be fairly cheap. And there’s nothing stopping countries from unilaterally deciding to spray their airspace, even though those materials would ultimately spread around the globe.
“I just have a hard time seeing with the economics of it how it doesn't happen,” “To me, that means that it's really urgent to do more research.”
Let's say you want to start geoengineering today to stabilize global temperatures where we are, or maybe bring it down a little bit. You basically need a fleet of airplanes that can reach the stratosphere. We're talking on the scale of maybe tens to hundreds of airplanes, and the capability to spray aerosol precursors.
But the way that the stratosphere works is that once you get up there, stratospheric winds take things around the planet relatively quickly in bands of latitude. And then slowly over time, on the timescale of months, things sort of migrate in general from equatorial regions up toward the poles, and then particles fall out near the poles.
So you wouldn't need to be flying through the whole stratosphere spraying stuff. The stratosphere does a lot of work to spread it out. And that's part of the reason why you can't restrict stratospheric geoengineering over just one area of the stratosphere.
"I'm having a hard time seeing how we're not going to do it at this point, actually, because it's so inexpensive. Already the impacts of climate change are looking to be so disruptive that I don't see in this world how such a low-expense solution doesn't get implemented by someone. There's just nothing else in the world that can cool the planet as quickly. Even if we started rapidly decarbonizing and taking CO2 out of the atmosphere, it's still a decade timescale for consequences. Whereas blocking sunlight, the climate response starts right away. "
If the program got disrupted, and if it was helping block a lot of warming, you would get this really rapid warming, it would be catastrophic
But the technology's not so complicated that we would need just the person who developed the technology to be the one to keep doing it. It's within the grasp of a medium-sized country.
The resources are substantial enough that it would be hard for a single individual, or a very small country to do it. But it's not like nuclear weapons or something like that.
From The Economist :
This would probably work (big volcanic eruptions, which do something similar, have a measurable effect on global temperatures).
The costs, though, could be considerable—and not just directly in dollars.
A poorly designed SAI programme might break down ozone, a form of oxygen that shields organisms, people included, from harmful ultraviolet radiation.
Patterns of precipitation would also change, for cooler air absorbs less moisture, and these effects would undoubtedly vary from region to region.
Another problem is the acid rain that would result.
Nor is the risk of someone doing something stupid a fantasy. In 2019 Massimo Tavoni, a game theorist at Milan Polytechnic who is unaffiliated with DARPA, organised six games played by 144 students. Participants were given a variety of ideal climate outcomes and allowed to spend toy money they were given on geoengineering projects to achieve them. Those who ended a round with their optimal climate outcome would then receive a payout of real cash. But an overabundance of SAI programmes were launched. To make matters worse, some players tried to counter efforts at cooling which they deemed excessive with attempts to warm the planet, resulting in a chaotic outcome that Dr Tavoni calls “geoengineering wars”. In the end, he says, “everybody loses.”
Les dernières newsletters :
L’addition ?
Cette newsletter est gratuite, si vous souhaitez m'encourager à continuer ce modeste travail de curation et de synthèse, vous pouvez prendre quelques secondes pour :
transférer cet email à un(e) ami(e)
étoiler cet email dans votre boîte mail
cliquer sur le coeur en bas d’email
Un grand merci d'avance ! 🙏
Ici pour s’inscrire et recevoir les prochains emails si on vous a transféré celui-ci.
Quelques mots sur le cuistot
J'ai écrit plus de 50 articles ces dernières années, à retrouver ici, dont une bonne partie publiés dans des médias comme le Journal du Net (mes chroniques ici), le Huffington Post, L'Express, Les Échos.
Retrouvez ici mon podcast Parlons Futur (ou taper "Parlons Futur" dans votre appli de podcast favorite), vous y trouverez entre autres des interviews et des résumés de livres (j’ai notamment pu mener un entretien avec Jacques Attali).
Je suis CEO et co-fondateur de l'agence digitale KRDS, nous avons des bureaux dans 6 pays entre la France et l'Asie. Je suis basé à Singapour (mon Linkedin, mon Twitter), également membre du think tank NXU.
Merci, et bon weekend !
Thomas