The biggest digital change since crypto, Steve Jobs ressuscité, changer de voix en 2 sec, IA crée une BD & plus
Vaccin contre le cancer en 2030 ? , l'évolution de l'humanité en images selon l'IA, robot pieuvre & plus
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
Un robot bat le record de vitesse de pliage de T-shirt (source)
can fold 30–40 randomly positioned garments per hour (2 minutes / garment) (see the 1-min video)
previous robotic garment-folding methods reached only 3-6 folds per hour
humans can generally do 120 items an hour, so we’re probably getting to human levels of folding within 1 or 2 years
Exemple de vidéo de 15 secondes éditée de façon stupéfiante et très facilement grâce à l'outil Runway utilisant l'IA, utilisant "erase and replace, slow mo and motion tracking." AI was used to animated face masks, generate jewelry and create rock formations. Et ici la vidéo making-of de 2 minutes, décortiquant les effets en images.
Dingue : autre démo vidéo de 40 sec de l'effet Erase and Replace de Runway (remplacer une coccinelle par un criquet dans une vidéo par ex, etc.)
Prototype allows users to listen to music through the sense of touch: "an algorithm that, by using "tactile illusions," renders monophonic music into tangible stimuli based on vibration, adding dynamics to the vibration in the form of movement, changes of direction and location." (source)
Cette techno permet de créer un site vitrine en 30 secondes top chrono (bon, très "templatisé", mais ça peut faire l'affaire) (source)
Ce robot utilise de nombreuses fines tentacules, comme une méduse, pour soulever délicatement des objets fragiles (voir la vidéo d'1min20)
The tentacles, or filaments, are made of rubber and aren't very strong individually. But working together, it can start to grapple some pretty heavy objects.
Since each individual tentacle doesn't grip strongly, even delicate cargo like plants can be ensnared without risking damage.
the researchers envision plenty of useful applications for the technology, like retrieving fragile artifacts on the ocean floor to moving fruits and vegetables in agricultural distribution.
Curieux: What happens when you ask an AI to generate "Human Evolution" in a succession of images : vidéo de 40 sec
Nouveau divertissement immersif à Las Vegas en 2023 : 20,000 spectateurs au centre d'une sphère de 130 mètres de diamètre tapissée entièrement du plus grand écran haute résolution au monde (source)
will be used for immersive, custom-made attractions, live performances, sports, gaming, and corporate events.
a shared experience: “You can put VR goggles on, and be transported to all these different places, but as you do that, firstly, you have no peripheral vision, and secondly, it isolates you. When you’re looking at a phone screen, you’re looking down, isolated. You’re not interacting with other people or with the shows that are happening. That’s where this is going to be very different."
there is the immersive sound suite, the 4D scent and air systems, the infrasound haptic vibrating seats. This is about putting all those pieces together and having it queued and timed properly to immerse people compellingly in the story
Progrès des graphismes dans les jeux vidéo : voyez cet extrait de 40 secondes "Amsterdam looks *incredibly* realistic in the new game Call of Duty — almost can’t believe this is a video game"
Now you can generate a "text prompt" by feeding it an image: latentspace.dev (la démo vidéo de 50 sec)
un "prompt" sert normalement à générer une image via des outils comme DALL-E et Stable Diffusion
AI called DeepStrike (by jabbr.ai) automatically analyses the performance of boxing opponents using camera feeds. (voir la démo vidéo de 35 sec)
The innovation could be used to replace boxing judges and provide athletes with accurate stats on their sparring training sessions to help them identify areas for improvement.
Et aussi ds le domaine du sport : New algorithms by the Cornell Laboratory can predict the in-game actions of volleyball players with more than 80% accuracy.
The algorithms are unique in that they take a holistic approach to action anticipation, combining visual data – for example, where an athlete is located on the court – with information that is more implicit, like an athlete’s specific role on the team.
One goal of the project is to help annotate game film, which is a tedious task when performed manually.
"I believe that this project has the potential to dramatically influence the way teams study and prepare for competition." said Ben Russell, director of hockey operations for the Cornell men’s team (see the video explaining what they did for volleyball and ice hockey)
Amusant : thismoviedoesnotexist.org uses AI (GPT-3 for text + StableDiffusion for images) to generate fake movie posters + title + synopsis.
Par ex : Merlin and Morgana : Merlin the magician must defeat the evil sorceress Morgana before she can take over Camelot and the surrounding kingdoms during the Middle Ages.
Wow : AI-generated picture of “a guy taking selfie during the battle of Omaha Beach, 1944” (source)
Wow : with a new AI tool called Koe Recast, you can transform up to 20 seconds of your voice into 10 different styles, including an anime character, a deep male narrator, an ASMR whisper.
assez dingue, enregistrez un message audio de qq secondes, même en français, choisissez une des voix, et juste qq secondes après écoutez le résultat après échange de voix
écoutez les exemples avec Mark Zuckerberg
Smart ! Musk just revealed how Tesla caught an employee who leaked trade secrets in 2008."We sent what appeared to be identical emails to all, but each was actually coded with either one or two spaces between sentences, forming a binary signature that identified the leaker".
"les 2 fondateurs de BioNTech -- les premiers à avoir réalisé un vaccin à ARN messager contre le Covid -- annoncent que "l’expérience acquise lors de la mise au point" dudit vaccin a permis de faire progresser rapidement la recherche si bien qu’un vaccin -- lui-aussi à ARN messager -- contre le cancer semble désormais réalisable "d’ici 2030". Les 2 scientifiques affirment même déjà disposer d’un "certain nombres d’avancées" prometteuses. (BBC via TTSO)
Parlons passé : The Autoped was an early motor scooter manufactured in New York from 1915 to 1921
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 (semaine prochaine, tout ce qu’il faut retenir des annonces de Facebook sur son metaverse + ce qu’ils mijotent à 5-10 ans, assez dingue)
À table !
Plus bas, 2 synthèses :
une sur le retour de Steve Jobs dans un podcast et ce que ça préfigure pour les techs du deuil,
et une autre plus laregement sur la nouvelle révolution (encore une), le vent de folie qui souffle depuis 6 mois dans la Silicon Valley autour de la “Generative Tech”
Incroyable: play.ht (text-to-voice AI startup) released a fake podcast featuring Steve Jobs talking with famous American podcaster Joe Rogan. The voices and the script were all generated by AI (source)
Les voix ne sonnent pas mal du tout, et c'est assez magique d'entendre Steve Jobs répondre à des questions plus ou moins inédites
"the Steve Jobs episode was trained on his biography and all recordings of him we could find online so the AI could accurately bring him back to life."
The 19-minute interview doesn't make much sense though. After a while, parts of the fictional interview begin to sound like conceptual mashups of common Jobs talking points, including aesthetics, revolutionary products, competitors such as Google, Microsoft, and Adobe, and the triumphs of the original Macintosh.
In this particular fictional world, Jobs is a huge Rogan fan. "It's nice to sit back in the car and listen to you rant," he says.
Cette techno devrait très bientôt se généraliser pour permettre de dialoguer avec des émulations de proches disparus, émulations audios, vidéos, tridimensionnelles en VR/AR, d'autant plus fidèles qu'il y aura d'informations en amont pour les "entraîner"
J'avais écrit déjà en 2015 sur ce sujet une tribune pas mal relayée à l'époque dans le média d'anticipation SingularityHub.com : Resurrecting the Dead—Bringing Back the Dearly Departed Digitally
La tendance a même un nom, "grief tech", la tech du deuil, extraits d'un article de la MIT Tech Review de ce mois-ci "Technology that lets us “speak” to our dead relatives has arrived. Are we ready?":
"From what I could glean over a dozen conversations with my virtually recreated parents, this really will make it easier to keep close the people we love," wrote MIT's Charlotte Jee, who, with the help of her living parents, tested out a grief tech startup called HereAfter AI for herself.
HereAfter’s interviewer spoke to each of my parent for several hours. The company then took those responses and started stitching them together to create the voice assistants. A couple of months later, a note popped into my inbox from Vlahos. My virtual parents were ready.
According to Jee, her AI "parents" first sounded "distant and tinny (à la voix un peu métallique)," but started to "sound more like themselves" over time. And it does seem like she got some value out of the experience, saying that she learned a few new stories about each of her parents' younger lives, heard some tales about her own childhood, and even got some life advice.
Coping with grief and loss is one of the most difficult parts of the human experience. But a crop of new startups like HereAfter — which asks users to provide hours of recordings based on specific prompts, then generates conversations from that dataset — seek to soften the blow.
While a photo album is static, an AI-generated voice that speaks from beyond arguably adds a different dimension of interactivity.
it’s possible to tweak large language models (LLMs) software like OpenAI’s GPT-3 or Google’s LaMDA to make it sound more like a specific person by feeding it lots of things that person said. In one example of this, journalist Jason Fagone wrote a story for the San Francisco Chronicle last year about a man who uploaded old texts and Facebook messages from his deceased fiancée to create a simulated chatbot version of her, using software known as Project December that was built on GPT-3. By almost any measure, it was a success: he sought, and found, comfort in the bot.
Users need to be careful not to think this technology is re-creating or even preserving people. “I didn’t want to bring back his clone, but his memory,” she says. The intention was to “create a digital monument where you can interact with that person, not in order to pretend they’re alive, but to hear about them, remember how they were, and be inspired by them again.”
In the first weeks and months after a loved one dies, people struggle to accept the loss and may find any reminders of the person triggering. “In the acute phase of grief, you can get a strong sense of unreality, not being able to accept they’re gone,” Thompson says. There’s a risk that this sort of intense grief could intersect with, or even cause, mental illness, especially if it’s constantly being fueled and prolonged by reminders of the person who’s passed away.
Un peu sinistre qd même, mais bon après tout c'est bientôt la Toussaint/Halloween au choix ...
Après le web 2 (social platforms), et les cryptos, on assiste depuis 6 mois on assiste à la naissance d'une nouvelle grande vague d'innovation startup : la Generative Tech
Extraits de la synthèse faite par le fonds d'investissement NFX, abondamment relayée, et qui relate bien le moment que l'on est en train de vivre :
The biggest change to the Internet since crypto just happened. A whole new world of applications opened up to Founders in the last 6 months.
We now have high-quality, cheap, fast AI models for generating text, images, videos, software code, music, voice, 3D models and more – none of which is copyrighted, and is not plagiarized.
The recent availability of open-source alternatives to proprietary generative AI models from Open AI is what caused it to tip wide open in the last 6 months. In short, Eleuther.ai’s GPT-NeoX-20B, launched Feb 2022, is the open source alternative to OpenAI’s GPT-3 for text generation. StabilityAI’s Stable Diffusion, launched August 2022, is the open source alternative to OpenAI’s DALL-E 2 for images and videos. Both have been game changers on price, quality and ease of access.
The cost to generate images has dropped 100X in the last 2 months.
Typically, major tech shifts roll out slowly. Many people were still skeptical of the Internet until 2003, so those of us who believed in it had less competition. Online software was gaining consensus from 1997 to 2005. Apple didn’t open their iOS platform to outside developers for 18 months after launch. Web3 has been rolling out for 10 years. But everyone is on board with Generative Tech. The investors get it. The Founders get it. The incumbents get it. And it’s clear that the game is now on.
Generative Tech is the next step in software. It’s a new level of human-machine partnership. It turns deep learning engines into collaborators to generate new content and ideas nearly like a human would.
The Generative Tech sector is developing at such a pace that it has already been validated by real revenues and high valuations, even though we collectively didn’t have a name for it until a month ago
Some have called it “Generative AI,” but AI is only half of the equation. AI models are the enabling base layers of the stack. The top layers will be thousands of applications. Generative Tech is about what will actually touch us – what you can do with AI as a partner.
For knowledge workers and creatives, going from zero to one in their minds will never be the same. For instance, writers, students, marketers, coders, architects, graphic designers, musicians, videographers, sales development reps, customer service reps, screenplay writers, and who are paid go from zero to something useful will now be using these tools to generate their first ideas.
Until now, software has been used to refine our initial ideas into something useful; it was responsible for the second half of the process, if you will, of going from zero to something useful. But these new generative tools help you with the first half of the process, taking you from nearly zero to a lot of initial ideas. And then the old software tools pick up from there and take you the rest of the way.
Until now, software couldn’t solve the zero to one problem because it worked FOR us. Generative Tech will work WITH us from the beginning of any project.
Sam Altman, CEO d'OpenAI dit que pour la première fois Google pourrait se faire challenger sur son coeur de métier
A clear example of the experience searching on Google vs asking GPT-3 (l'IA génératrice de texte d'OpenAI).
One example : The.com allows you to generate hundreds of new websites all within one spreadsheet. These aren’t cookie cutter copies, they’re beautifully designed, and unique to the needs of each user.
An example from the biology world is Insilico Medicine: it uses 3 AI-powered products that work together.
One identifies new targets for drugs,
while another generates new candidate molecules from scratch.
Finally, the last engine predicts the outcome of clinical trials based on previous work.
Infographie du fonds d’investissement Sequoia faisant listant les principales 120 startups de la Generative Tech à ce jour, divisée en ces catégories : Text, Image, Video, Code, Speech, 3D, Music, Gaming, Biology, Others (et ici encore une autre liste de startups classées avec liens)
Extrait d’un article du New York Times (derrière paywall) sur Stability.ai, l’entreprise derrière l’IA génératrice d'images Stable Diffusion, qui vient de lever 100m$
the computer the company uses to train its A.I. models, that has more than 5,000 high-powered graphics cards and is already one of the largest supercomputers in the world, would grow to five or 10 times its current size within the next year.
that firepower would allow the company to expand beyond A.I.-generated images into video, audio and other formats, as well as make it easy for users around the world to operate their own, localized versions of its algorithms
Emad Mostaque, the founder and chief executive of Stability AI, has pushed back on the idea of content restrictions. He argues that radical freedom is necessary to achieve his vision of a democratized A.I. that is untethered from corporate influence.
Mr. Mostaque, 39, is an odd frontman for the generative A.I. industry. He has no Ph.D. in artificial intelligence, nor has he worked at any of the big tech companies from which A.I. projects typically emerge, like Google or OpenAI. He is a British former hedge fund manager who spent much of the past decade trading oil and advising companies and governments on Middle East strategy and the threat of Islamic extremism. More recently, he organized an alliance of think tanks and technology groups that tried to use big data to help governments make better decisions about Covid-19.
Mr. Mostaque, who initially funded Stability AI himself, has quickly become a polarizing figure within the A.I. community. Researchers and executives at larger and more conventional A.I. organizations characterize his open-source approach as either naïve or reckless. Some worry that releasing open-source generative A.I. models without guardrails could provoke a backlash among regulators and the general public that could damage the entire industry.
But, on Monday night, Mr. Mostaque got a hero’s welcome from a crowd of several hundred A.I. researchers at this week’s Stability AI launch event in San Francisco.
Démo vidéo de 2 minutes de Lex, tout nouvel outil d'aide à l'écriture grâce à l'IA : "This is surprisingly good"
Ce que le founder de Jasper.ai (autre outil d'écriture via IA , "Create amazing text content 10X faster with AI") : "In AI, the models are a commodity in the future. No moat (barrière à l'entrée) there. One great moat is in the data you use to train and fine-tune the models."
Certains artistes perplexes face à cette révolution, extraits d'un article de Business Insider qui a bien tourné sur Twitter :
Artists say AI image generators are copying their style to make thousands of new images — and it's completely out of their control
"We work for years on our portfolio," Rutkowski said. "Now suddenly someone can produce tons of images with these generators and sign them with our name."
Rutkowski's name has been used to generate around 93,000 AI images on one image generator, Stable Diffusion — making him a far more popular search term than Picasso, Leonardo Da Vinci, and Vincent van Gogh in the program.
"I feel like something's happening that I can't control," Rutkowski, who is based in Poland, told Insider. "My name is being used a lot to generate AI images, along with the names of other working artists."
Giles Christopher, a London-based commercial photographer specializing in food and drink, uses DALL-E and other AI-image generators to experiment with portraits and create artificial backgrounds for some of his commercial shots.
"I've come out with images that you wouldn't question are photographs," he said. "Some of the arguments I've had from photographers are that the images are looking too good."
He thinks the genie is out of the bottle when it comes to AI, and that artists should look for ways to include it into their work
Et les premières réactions dans le monde de la musique : "the Recording Industry Association of America listed AI-powered music websites that make remixes, improve homemade tracks, or strip songs of vocals or instrumentals harm artists" (source)
"In any event, the files these services disseminate are either unauthorized copies or unauthorized derivative works of our members’ music.”"
The RIAA specifically names Acapella-extractor and Remove-Vocals, which remove the instrumentals or the vocals from songs, respectively, and Songmastr, which claims to “make your songs (almost) as good as” something made by a professional artist, by using their music as a reference.
Combat perdu d'avance ?
15 areas that DALL-E 2 and image-generation AI are already disrupting
Stock photos, Graphics novels, Children's books, Game characters and graphics, E-commerce product ideas, Newsjacking, Logo designs, Marketing, Clothing designs, Tattoos, Photo shooting and editing, Interior design, Content writing
Graphics novels: A graphic novel entirely generated with MidJourney, a DALLE-2 competitor which ended up on the Amazon's Best Seller list
Children's books: Some people have managed to make children's books in 2 hours with text-to-image AI, including research, image generation and layout design.
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