🚀Why Google Missed ChatGPT, Is fusion power 2 or 20 years away?, ChatGPT just killed take-home exams: 3 solutions, & more
Vaccin contre le cancer ? ; ce robot gère 50% des appels à cette mairie, & 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
L'entreprise chinoise Xioami conçoit un robot humanoïde qui peut jouer de la batterie
"The first reason Xiaomi is working on humanoid robots is that we are seeing a huge decline in the labor force in China, and the world." (source)
Sérieusement ? Dyson has a weird $1k battery-powered headphone/face mask/air purifier (source)
"can capture up to 99% of the particle pollution, and carbon filters target gasses associated with city pollution, such as nitrogen dioxide and sulfur dioxide."
targets the Chinese market first and foremost
"The air filtration system can be detached from the headphones" (encore heureux)
More crazy ChatGPT stories
Bluffant : ce médecin fait la démo vidéo en 2 minutes de comment il fait rédiger par l'IA une lettre éloquente avec références scientifiques pour convaincre une compagnie d'assurance de couvrir une dépense de patient souvent rejetée, simplement en demandant : "Write a letter to United healthcare asking them to approve an Echocardiogram on a patient with Systemic Sclerosis. Make references to supporting scientific literature. List the appropriate articles."
Someone got ChatGPT to generate a: 💡Movie Concept, 🗺️ Plot, 🤖 Character Bio and 🎬 Sequel, all in 11 messages, and then Stable Diffusion to "design" 🖌️ Concept Art for it (source)
Someone, not particularly tech-minded, wrote a half-decent Excel tool with it that saved him a few hours at work” (source)
He works in higher education, studied English at university, and never formally learned to code.
But here he was, not only playing around with an experimental AI chatbot but using it to do his job faster after only a few days’ access.
“I asked it some questions, asked it some more, put it into Excel, then did some debugging,” is how he described the process. “It wasn’t perfect but it was easier than Googling.”
An author showing how ChatGPT could completely disrupt their industry (source with actual examples)
I put together a demo in 2 minutes, showing that ChatGPT can come up with ideas of novels in seconds across a range of genres.
ChatGPT can come up with your main characters, your storyline, and more for you, all with incredible ease
Finally, if you can't be bothered actually writing the thing, ChatGPT will get started for you. It does a surprisingly good job of this, and no doubt the technology will continue to improve.
It will even compose an e-mail to an agent for you pitching the book. This is science fiction, folks. And, I haven't even scratched the surface when it comes to writing scenes.
He shares more examples of ChatGPT writing engaging fiction in real-time, without pausing to think or having to research the topic. Here, it tackles historical fiction and crime fiction, two completely different genres, with ease. And remember, this is just the beta version.
"I spent the weekend playing with ChatGPT, MidJourney, and other AI tools… and by combining all of them, published a children’s book co-written and illustrated by AI!" (le détail)
Read the answer to the prompt: "Write a biblical verse explaining how to remove a peanut butter sandwich from a VCR (magnétoscope)"
The bones of ChatGPT are not entirely new (it’s based on GPT-3.5, a large language model that was released by OpenAI this year but which itself is an upgrade to GPT-3, from 2020). OpenAI has previously sold access to GPT-3 as an API, but the company’s ability to improve the model’s ability to talk in natural dialogue and then publish it on the web for anyone to play with brought it to a much bigger audience. And no matter how imaginative AI researchers are in probing a model’s skills and weaknesses, they’ll never be able to match the mass and chaotic intelligence of the internet at large.
🤯 OpenAI is rumored to launch GPT-4 as soon as in 2023, with 500 times more parameters than GPT-3, a proxy for how much better it may work
Débuts des tests pour un vaccin contre le cancer de la peau par Moderna (TTSO)
1ère fois dans l’histoire de la médecine, le labo US Moderna va débuter la phase 3 (= tests sur l’homme à grande échelle) d’un vaccin contre le cancer (en l'occurrence celui de la peau).
Une avancée majeure qui s'explique par les progrès de l'ARN messager, une technologie connue depuis 30 ans mais boostée par le développement des vaccins anti-Covid.
Et les résultats sont ultra prometteurs : la phase 2 réalisée sur 147 patients a montré une baisse de 44% du risque de rechute ou de décès
Selon le CEO de Moderna, ce nouveau vaccin devrait même être efficace contre d’autres formes de cancers.
This tool sumup.page will summarize an article, just enter a link. Very easy to test, free, no registration
TSMC, le fabricant taiwanais de semiconducteurs, investit 40 milliards de $ aux Etats-Unis pour y construire des usines (montant du même montant que la plan de la Commission Europénne)
TSMC produit 90% des puces haut de gamme (le reste par le Coréen Samsung)
Alors, un moyen pour les Etats-Unis de protéger son accès à la technologie en cas d'annexion de Taiwan par la Chine ?
Pas complètement, TSMC conserve à Taiwan le très haut de gamme, les puces de 1 ou 2 nanomètres, et si 500 ingénieurs partent aux Etats-Unis, il en reste 50,000 à Taiwan, où TSMC vient d'annoncer la construction d'un usine d'1 nanomètre, le graal de cette industrie.
(Source: Pierre Haski, sur France Inter)
Et comme Thomas Friedman écrivait dans le NYT l'année dernière "If China thinks it can get around that by seizing Taiwan just to get hold of TSMC, that would be a fool's errand. Many of the key machines and chemicals TSMC uses to make chips are from America and the European Union, and that flow would immediately be shut down. (voir le résumé de son article dans cette précédente newsletter, cherchez "TSMC" pour retrouver le passage)
Le Youtubeur américain Tim Dodd, connu sous le pseudo The Everyday Astronaut pour ses vidéos de vulgarisations, a été choisi par le milliardaire japonais Yusaku Maezawa pour l'accompagner, avec 7 autres créateurs, autour de la lune dans quelques années à bord d'un Starship de SpaceX, bien mérité, sa vidéo pleine d'émotions vaut le détour pour les passionné d'espace
Le "callbot" de la startup Yelda répond à 100% des appels entrants à la mairie de Plaisir, commune des Yvelines
permet un taux de décrochage de 100%
permet de réduire jusqu'à 50% le nombre d'appels gérés par les standardistes
des appels de 6 à 7 minutes maintenant gérés entièrement par le callbot
des réponses plus rapides pour les usagers, des standardistes qui passent plus de temps sur les demandes qui le nécessitent
écouter le témoignage de la chargée de mission qualité au sein de la direction de la relation citoyenne de la mairie
"pas besoin d'avoir fait BAC+10 pour prendre en main le logiciel", dit-elle
(par transparence, startup lancée par 2 de mes associés)
Pour Noël, faites plaisir aux curieuses et curieux dans votre entourage ❤️
Si vous appréciez cette synthèse, prenez s’il vous plaît 3 secondes pour “offrir” un abonnement à ne serait-ce qu’une personne (ou 2, ou 3…) 🙂
C’est gratuit, il/elle pourra toujours se désabonner d’un clic !
À table !
Tesla en Europe, ça se complique : Tesla’s Berlin Hub Can’t Hire Enough People, or Keep Them (Wired)
according to former and current employees at the gigafactory, current staffers are leaving jobs due to low and unequal pay and inexperienced management in the highly competitive German manufacturing sector
One current employee, who requested anonymity out of fear of losing their job, describes the Berlin gigafactory as “total chaos.” “Some people are off sick longer than they’ve actually worked. There are people who I haven’t seen working for three weeks in six months.
People in HR want to hit their targets for recruitment, so they will say anything to get people in, but not pay attention to keeping these workers,” they say.
"unannounced changes in working conditions."
Tesla’s attempt to improve recruitment and retention by increasing pay for new staff also backfired, as longer-term employees were being paid less than employees who had just arrived doing the same jobs with similar qualifications.
A female tech writer tested the latest viral AI app Lensa that lets you create profile pictures from a few selfies (source)
"I was hoping to get results similar to my colleagues at MIT Technology Review. The app generated realistic and flattering avatars for them—think astronauts, warriors, and electronic music album covers."
"Instead, I got tons of nudes. Out of 100 avatars I generated, 16 were topless, and another 14 had me in extremely skimpy (short and revealing) clothes and overtly sexualized poses."
Lensa generates its avatars using Stable Diffusion, an open-source AI model that generates images based on text prompts. Stable Diffusion is built using a massive open-source data set that has been compiled by scraping images off the internet.
And because the internet is overflowing with images of naked or barely dressed women, and pictures reflecting sexist, racist stereotypes, the data set is also skewed toward these kinds of images
The company says that because Stable Diffusion is trained on unfiltered data from across the internet, neither they nor Stability.AI, the company behind Stable Diffusion, “could consciously apply any representation biases or intentionally integrate conventional beauty elements.”
ChatGPT just killed homework, essays, and take-home exams: 3 paths forward (source)
Apps to answer math questions have been around for a while.
But ChatGPT is different - it can do things that previously required human judgment and analysis, like writing full essays or solving complex problem sets.
The result? An "existential crisis" for educators.
3 paths forward:
1. Schools adjust assignments to prevent the use of AI.
Take-home work largely disappears. Class time is used for proctored essays, problem sets, and exams.
Homework time is spent learning asynchronously via video - a "flipped classroom" model.
(this is the method used by the sucessful online Khan Academy for years)
2. Schools embrace AI.
Students will use AI in real life. Why make them do things the "old fashioned way" at school?
Instead, lessons will incorporate AI - teaching students how to write prompts, analyze outputs, and edit as needed
3. Schools learn to audit AI.
In this case, AI assistance is viewed like plagiarism. Educators learn how to detect it, and have policies in place to downgrade or disqualify assignments.
A "GPT watermark" may already be in the works at OpenAI
An investor tweeted (and then deleted)
Society is not ready for a step function inflection in our ability to costlessly produce anything connected to text. Human morality does not move that quickly, and it’s not obvious how we now catch up.
I am hearing from colleges unsure what to do about grading ALL ESSAYS this term. Hundreds and hundreds of essays. The examples are legion across many, many domains. A societal trust collapse, at scale. I obviously feel ChatGPT (and its ilk) should be withdrawn immediately.
2 examples of how Artificial intelligence is permeating business at last (The Economist)
agriculture
This autumn John Deere, a tractor-maker, shipped its first fleet of fully self-driving machines to farmers. The tilling tractors are equipped with six cameras which use artificial intelligence (AI) to recognise obstacles and manoeuvre out of the way.
50% of the vehicles John Deere sells have some AI capabilities. That includes systems which use onboard cameras to detect weeds among the crops and then spray pesticides, and combine harvesters (moissonneuses-batteuses) which automatically alter their own setting to waste as little grain as possible.
For a medium-sized farm, the additional cost of buying an AI-enhanced tractor is recouped in 2 to 3 years.
programming
A number of firms are offering a virtual assistant trained on a large deposit of code that churns out new lines when prompted.
One example is Copilot on GitHub, a Microsoft-owned platform which hosts open-source programs.
Programmers using Copilot outsource nearly 40% of the code-writing to it. This speeds up programming by 50%, the firm claims.
Nouveau design d'éolienne flottante par une startup française (source)
Wind energy’s pitfall is that it’s expensive, particularly offshore. Some studies have gone so far as to proclaim offshore wind’s “dismal economics” unsustainable, particularly without massive government subsidies to keep it afloat.
French startup Eolink is building a five-megawatt floating turbine (enough to power 5,000 homes)
The floating part matters because else transporting, installing, and maintaining an enormously tall and heavy turbine in the ocean depths is costly and difficult.
What if the weight could instead be distributed between multiple support poles? That’s the design Eolink has in mind; its floating turbine swaps out the single large pole for four thinner ones angled towards each other in a pyramid shape. This not only distributes the weight of the turbine’s pieces, it allows the whole structure to be lighter.
The company says its design will use about 33% less steel than a traditional turbine with the same capacity, and thanks to a longer distance between the blades and mast, could produce 10% more energy at equivalent wind speeds.
They hope to eventually scale up to a 20-megawatt turbine with the same design. (enough to power 20,000 homes)
Why Google Missed ChatGPT (source)
After years of preaching that conversational search was its future, it’s stood by as the world discovered ChatGPT.
This was the future Google promised. But not with someone else fulfilling it.
“Google thinks a lot about how something can damage its reputation,” said Gaurav Nemade, an ex-Google product manager who was first to helm its LaMDA chatbot. “They lean on the side of conservatism.”
Google’s LaMDA — made famous when engineer Blake Lemoine called it sentient earlier in June 2022— is a more capable bot than ChatGPT, yet the company’s been hesitant to make it public.
For Google, the problem with chatbots is they’re wrong a lot, yet present their answers with undeserved confidence. Leading people astray — with assuredness — is less than ideal for a company built on helping you find the right answers. So LaMDA remains in research mode.
On the one hand, a sophisticated, public chatbot like ChatGPT makes waiting for the perfect business model risky. Delay long enough, and you could cede the market.
On the other hand, ChatGPT will also take criticism as it gains adoption, sustaining hits that otherwise would’ve been Google’s. ChatGPT’s shortcomings will teach people to view its certainly with skepticism, clearing the way for a risk-averse Google to release its own version.
For now, ChatGPT’s threat to Google remains somewhat limited. The bot doesn’t access the internet, knows nothing beyond 2021 (or at least, it says so), costs a few cents per chat to OpenAI, and has no ads. So while it may take some traditional queries away from Google, it won’t push the $1.2 trillion company to the brink. At least as presently constituted.
But things could change in a hurry. Should OpenAI connect ChatGPT to the internet, it could push Google to bring its own product to market, and its vision for the future along with it. And once Google gets involved, those who’ve seen its chatbot technology expect it to win.
“If ChatGPT or some other product ever became a real threat,” said Lemoine, “they'd just bite the bullet and release LaMDA, which would smoke ChatGPT.”
What's the deal with fusion power? Decades away, or 2 years away?
Nuclear fusion is a man-made process that replicates the same energy that powers the sun. Nuclear fusion happens when two or more atoms are fused into one larger one, a process that generates a massive amount of energy as heat. That heat can be used to warm water, create steam and turn turbines to generate power – much like how nuclear fission generates energy.
Scientists around the world have been studying nuclear fusion for decades, hoping to recreate it with a new source that provides limitless, carbon-free energy – without the long-lived nuclear waste created by current nuclear reactors. Fusion projects mainly use the elements deuterium and tritium – both of which are isotopes of hydrogen
Hydrogen isotopes are atoms of hydrogen with different numbers of neutrons in their nuclei (neutrons in blue above), respectively 1 and 2 neutrons for deuterium and tritium, in addition to the only one proton that defines the hydrogen atom (protons in red above)
The deuterium from a glass of water, with a little tritium added, could power a house for a year. Tritium is rarer and more challenging to obtain, although it can be synthetically made. A small cup of the hydrogen fuel could theoretically power a house for hundreds of years.
Physicists have since the 1950s sought to harness the fusion reaction that powers the sun, but no group had been able to produce more energy from the reaction than it consumes.
On December 5, researchers at the Lawrence Livermore Lab’s National Ignition Facility (NIF) finally took the perfect shot: this is the first time scientists have ever successfully produced a nuclear fusion reaction resulting in a net energy gain, instead of breaking even as in past experiments.
2 megajoules in (the energy used in 15 minutes of running a hair dryer, but delivered all at once, in a millionth of a second) & 3 megajoules out. A 50% gain of energy.
The difference is in how scientists define breakeven. Today, the NIF researchers said they got more energy out than their laser fired at the experiment
But the problem is that the energy in those lasers represents a tiny fraction of the 200 megajoules involved in firing up those lasers. By that measure, NIF is getting way less than they’re putting in. “The real type of breakeven is way, way, way, way down the road,” Cappelli says. “That’s decades down the road. Maybe even a half century down the road.”
Generating fusion energy using NIF’s method involves shooting dozens of laser beams into a gold cylinder, heating it up to more than 3 million °C. The lasers don’t target the fuel directly. Instead, their aim is to generate “a soup of x-rays.” These bombard the tiny pea-sized (de la taille d'un petit pois) fuel pellet consisting of the hydrogen isotopes deuterium and tritium, and crush it.
The compression caused by the beams (or, rather, by x-rays generated by their effect on the capsule containing the pellet) overcomes the mutual electrical repulsion of the nuclei of the atoms in the pellet, which are all positively charged. It thereby pushes those nuclei close enough to one another for a different fundamental force, the strong nuclear force (which operates only at short ranges) to take over.
The trouble is inefficient lasers. Producing those laser beams at NIF involves a space nearly the size of a football field built at a cost $3.5 billion, filled with flashing lamps that excite the laser rods and propagate the beams. That alone takes 300 megajoules of energy, most of which is lost.
Add to that layers of cooling systems and computers, and you quickly get an energy input that’s multiple orders of magnitude greater than the energy produced by fusion.
To produce sustained energy, scientists need to figure out how to repeatedly fire the powerful lasers and get many pellets in front of them. That could involve multiple pellets and laser firings per minute. By comparison, NIF currently fires 3 times per day.
This type of fusion experiment with laser beams is known as “inertial confinement", it has gotten less attention than another type of fusion technology known as “magnetic confinement” which involves a donut-shaped device known as a tokamak, in which hydrogen gas is heated into plasma and then trapped by magnetic fields. Commercial fusion companies have generally taken the magnet route, as has ITER, in part because of the challenges of lasers. But recently, inertial facilities have seen more investment—and today’s success may mean more of that ahead.
The Economist is harsher: It seems unlikely that the future of civil fusion power (if it has one) lies with inertial-confinement by laser. The technology is fiddly. And even with lasers more modern than that used by NIF (which opened in 2009) the process of “pumping” the device to create the beam is inherently inefficient. None of the increasingly numerous attempts to commercialise fusion employs inertial-confinement by laser. Most are based on tokamaks.
Despite today’s announcement, fusion is neither commercial nor close to commercial, so it is still vaporware,” says Mark Jacobson, an energy researcher at Stanford who has argued for more investment in available solutions like solar, wind, and hydropower. Indeed, you would be hard-pressed to find a plasma physicist who thinks fusion will be in the mix in the next decade.
Sources: Wired, The Economist.
But...now we have a way to produce fusion “fire” on demand. The knowledge we will gain from this capability will greatly benefit all other fusion efforts
Pour rappel :
Sam Altman, CEO d'OpenAI, a investi $375 million dans Helion, "his largest investment in a start-up ever":
“by far the most promising approach to fusion I’ve ever seen,” said Altman in November 2021. “With a tiny fraction of the money spent on other fusion efforts, and the culture of a startup, this team has a clear path to net electricity. If Helion is successful, we can avert climate disaster and provide a much better quality of life for people.”
July 2022 update, Sam Altman said:
"Helion has been progressing even faster than I expected and is on pace in 2024 to 1) demonstrate fusion with a net energy gain and 2) resolve all questions needed to design a mass-producible fusion generator."
"The goals of the company are quite ambitious—clean, continuous energy for 1 cent per kilowatt-hour, and the ability to manufacture enough power plants to satisfy the current electrical demand of earth in a ten year period." (en France le kilowatt-heure coûte environ 15 centimes pour le consommateur final)
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