L'algo de X / X's algorithm
🇫🇷 Le meilleur compte de X avant Elon Musk ? Il n'existait pas encore ! 😎
🇬🇧 The best X account before Elon Musk? It didn't exist yet! 😎
L'algo de X / X's algorithm
Going public does not guarantee survival. Not even for an artificial intelligence company worth billions.
Anthropic's stock market debut raises a question that deserves more attention: how will AI companies ensure their long-term financial survival?
An IPO provides access to capital. It does not guarantee profitability, unlimited funding, or protection against bankruptcy.
And developing artificial intelligence requires enormous, continuous investments in computing power, infrastructure, energy, and research.
This creates a powerful economic incentive: recurring revenue.
Subscriptions, usage-based pricing, additional tokens, premium features, and long-term contracts.
We have already seen similar business models in the automotive industry, where features such as advanced driving assistance can be offered through subscriptions rather than permanent purchases.
But artificial intelligence could take this economic model much further.
AI is gradually moving beyond chatbots.
It is becoming an intermediary between humans and the technologies they use every day.
Our computers, smartphones, vehicles, homes, workplaces, banking services, healthcare systems, and administrative procedures could increasingly be connected through AI agents.
Consider the transformation.
Today, humans access digital services directly.
Tomorrow, they may increasingly access those same services through an AI intermediary.
And whoever controls that intermediary could influence which services we use, what information reaches us, and how much we pay.
The economic opportunity is enormous.
But so is the potential dependency.
A subscription may begin as a convenient choice and gradually become difficult to avoid when essential activities depend on the same technological infrastructure.
None of this guarantees that AI companies will survive. Recurring revenue does not automatically mean profitability, and consumers can reject restrictive business models.
Yet the underlying economic incentive is clear.
The more activities AI connects, the more opportunities companies have to monetize access to those activities.
Which brings us to a fundamental question:
Are AI companies competing to develop intelligence, or to become the indispensable paid intermediary between human beings and their everyday lives?
#AI #Anthropic #OpenAI
A human player learns from the virtual world.
An AI agent can also update its behavior through interaction.
Both processes involve information.
But their biological and computational mechanisms are different.
Similar performance does not establish identical intelligence.
The United States does not have a monopoly on artificial intelligence. But it has something that may prove just as powerful: the ability to finance it.
Anthropic has entered the stock market. Chinese AI companies such as Zhipu AI and MiniMax have also gone public. Europe has Mistral AI, India is developing its own AI ecosystem, and Russia continues investing in domestic technologies.
So why does the United States attract such extraordinary amounts of capital?
The answer may have less to do with intelligence itself than with the financial infrastructure surrounding it.
American companies benefit from deep capital markets, global investors, powerful cloud providers, and an ecosystem capable of turning technological expectations into billions of dollars in funding.
That capital finances computing power, researchers, infrastructure, and successive generations of AI models.
But here is the distinction we must not overlook.
Financial superiority is not scientific superiority.
A company's valuation measures investors' expectations, not the intelligence of its technology.
And spending more money does not automatically produce a proportionally more intelligent system.
The United States may therefore possess a decisive financial advantage without necessarily holding a permanent scientific advantage.
Which raises a fundamental question:
Are we witnessing a global race to understand intelligence, or a race to control the financial infrastructure that commercializes it?
#AI #Anthropic #OpenAI
Could a future GTA evolve without its developers scripting every event?
Adaptive systems could modify traffic, economies or character interactions.
But their behavior would still depend on architecture, objectives and constraints.
Autonomy does not mean independence from the system that makes it possible.
An AI character could accumulate experiences during thousands of simulated days.
Its behavior could change as its stored information changes.
But accumulated information alone is not intelligence.
The question is how a system uses previous states to adapt to new situations.
Imagine a simulated city where characters move, trade and cooperate even when no human is playing.
The city could maintain an evolving internal state.
But autonomous activity would not prove consciousness.
The first scientific question would be whether the simulated system follows coherent causal rules.
Today, a screen separates the player from the virtual world.
Tomorrow, spatial interfaces may reduce that separation.
But increased immersion does not automatically produce better understanding.
A system can stimulate more senses while providing less useful information.
More sensory input is not always more intelligence.
Imagine GTA in 2036.
A virtual city might be projected into an augmented-reality environment.
Real streets and digital objects could coexist in the player's field of view.
The technical challenge would be spatial alignment.
The cognitive challenge would be distinguishing physical reality from simulation.
Imagine a mission that was never written by a developer.
An AI generates the situation, characters and possible objectives.
But a coherent mission still requires constraints, causal relationships and verifiable outcomes.
Generating possibilities is easier than guaranteeing a consistent world.
What if open-source AI is playing a longer game than the companies trying to control access to intelligence?
In chess, winning material does not necessarily mean winning the game.
A player may sacrifice an immediate advantage to gain mobility, control the board, and create opportunities that only become visible several moves later.
The same strategic principle may apply to artificial intelligence.
Today, proprietary AI companies invest billions in computing infrastructure, research, and increasingly powerful models. Their economic survival depends on generating revenue through subscriptions, APIs, and commercial services.
Meanwhile, open-source and open-weight AI ecosystems continue to expand.
Their strategic advantage is different.
When technology becomes accessible, thousands of developers, researchers, and independent organizations can experiment with it, adapt it, identify weaknesses, and explore alternatives.
Innovation is no longer confined to the laboratory that created the original model.
And here is where biological evolution offers an interesting analogy.
Throughout evolution, organisms have exploited accessible resources and opportunities within their environments.
Nature does not operate according to economic concepts of ownership or subscriptions. But adaptation depends partly on the ability to exploit available resources effectively.
Human intelligence has followed another remarkable path: accumulating, transmitting, and reusing knowledge across generations.
Knowledge becomes especially powerful when it can be shared, tested, challenged, and improved.
Of course, open source is not necessarily free of costs, and openness alone does not guarantee technological superiority.
But consider the long-term strategy.
A proprietary company can control access to its technology.
An open ecosystem can multiply the number of minds working to improve it.
One strategy concentrates resources.
The other can distribute experimentation.
And technological history reminds us that the most financially powerful architecture does not necessarily remain dominant forever.
The decisive move in AI may not be controlling who can use intelligence, but enabling more people to develop it.
So here is my question:
Are proprietary AI companies winning today's commercial battle while open-source ecosystems are quietly building a stronger position for tomorrow?
In chess, the value of a move is not determined only by what it captures today, but by the possibilities it creates for the rest of the game.
#OpenSource #AI #Chess
What if GTA had a personal AI agent?
Not an automatic player.
An assistant capable of observing situations, explaining possible strategies and helping the player understand consequences.
The objective would not be to replace human intelligence.
It would be to extend the information available to it.
Imagine returning to a neighborhood after twenty hours of gameplay.
An NPC remembers your previous actions. Other characters react to their consequences.
The environment has acquired a history.
The scientific challenge would be maintaining consistency across interactions, not simply generating new dialogue.
An AI character might generate thousands of different sentences.
But producing language is not the same as understanding the physical consequences of an action.
A virtual character that says "I'm afraid" does not establish that it experiences fear.
Simulation and biological experience must remain distinct.
Imagine an NPC that remembers every encounter with the player.
Not merely a scripted response, but an evolving record that influences future interactions.
That would require more than convincing dialogue.
It would require memory, state management and consistent behavioral adaptation.
A player can break traffic rules, ignore a mission or explore an unfamiliar street.
The game's possibilities remain constrained by its architecture.
Yet within those constraints, human decisions can be unpredictable.
Freedom inside a simulation is not the same as freedom from its rules.
Et si le véritable avenir de l'intelligence artificielle n'était pas de remplacer Homo sapiens, mais de lui apprendre tout ce qu'il souhaite apprendre ?
Depuis l'arrivée de ChatGPT, nous parlons principalement de performances, de concurrence avec l'intelligence humaine, de disparition des emplois et de machines capables de nous surpasser.
Mais nous passons peut-être à côté de l'un des plus grands marchés économiques de l'histoire de l'apprentissage humain.
Imaginons un instant.
Vous souhaitez apprendre le français, les mathématiques, la physique, la musique, la peinture, la cuisine, la mécanique ou une langue étrangère.
Aujourd'hui, vous devez généralement trouver un enseignant, acheter des supports pédagogiques, suivre une formation ou consulter différentes ressources.
Demain, votre professeur pourrait être une intelligence artificielle.
Et pas simplement un chatbot qui répond à vos questions.
Un véritable professeur personnel, disponible à tout moment, capable de vous accompagner dans votre environnement réel.
Vous souhaitez apprendre le français ?
Votre professeur IA adapte ses exercices à votre niveau, identifie vos erreurs, explique les règles grammaticales, vous fait travailler la prononciation et vérifie progressivement vos acquis.
Vous souhaitez apprendre la guitare ?
Il vous écoute jouer, analyse votre rythme, identifie vos erreurs et vous propose des exercices adaptés.
Vous souhaitez apprendre à cuisiner ?
Il observe votre préparation à travers une caméra, vous guide vocalement, explique les réactions physiques et chimiques des ingrédients et corrige vos gestes lorsque cela est possible.
Vous préparez votre permis de conduire ?
Il vous enseigne le Code, simule des situations de circulation en réalité virtuelle et vous explique vos erreurs avant votre apprentissage pratique sur route.
Et pourquoi se limiter à un écran ?
L'intelligence artificielle pourrait prendre plusieurs formes complémentaires.
Un professeur exclusivement vocal.
Un enseignant virtuel présent dans des lunettes de réalité augmentée.
Un robot humanoïde capable de réaliser physiquement une démonstration.
Ou simplement un assistant intégré à votre téléphone, vous accompagnant tout au long de votre apprentissage.
La technologie ne serait plus seulement un système auquel nous demandons des réponses.
Elle deviendrait un environnement pédagogique interactif.
Et c'est ici que le potentiel économique devient considérable.
Car l'être humain n'apprend pas uniquement pendant son enfance ou sa scolarité.
Il apprend pendant toute son existence.
À chaque nouvelle activité, chaque nouvelle technologie, chaque changement professionnel ou chaque compétence qu'il souhaite acquérir.
Autrement dit, le marché potentiel ne se limite pas aux étudiants.
Il concerne pratiquement toute la population humaine, à tous les âges et dans presque toutes les disciplines.
Abonnements éducatifs, enseignement personnalisé, simulations immersives, robots pédagogiques, formations professionnelles et accompagnement technique.
Une multitude de services pourraient être proposés avec des revenus récurrents.
Nous retrouvons alors notre interrogation sur la survie financière des entreprises d'intelligence artificielle.
Une introduction en Bourse ne garantit ni la rentabilité ni la survie d'une entreprise.
Mais transformer l'IA en infrastructure mondiale d'apprentissage pourrait ouvrir des perspectives économiques considérables.
Attention cependant à ne pas confondre la capacité de fournir des explications avec celle de transmettre réellement une compétence.
Un professeur doit pouvoir vérifier les acquis, identifier les erreurs et tenir compte des contraintes de la réalité physique. L'IA ne possède pas encore toutes ces capacités avec la fiabilité nécessaire dans chaque discipline.
L'enseignant humain, l'expérience pratique et les validations professionnelles conserveront donc des fonctions essentielles.
Mais réfléchissons à la transformation fondamentale qui pourrait se produire.
Jusqu'à présent, une grande partie de l'économie numérique consistait à vendre des informations, des contenus et des outils.
Demain, une part croissante de cette économie pourrait consister à vendre un accompagnement personnalisé pour apprendre à utiliser ces informations et ces outils.
Et nous pourrions assister à un paradoxe fascinant.
Pendant que certains redoutent une intelligence artificielle qui remplacerait les capacités humaines, d'autres pourraient utiliser cette même technologie pour développer leurs propres capacités.
L'intelligence artificielle n'aurait donc pas besoin de devenir supérieure à l'intelligence humaine pour constituer un immense marché.
Il lui suffirait de devenir suffisamment performante pour accompagner l'apprentissage humain dans un nombre croissant de situations.
Et c'est peut-être là que réside l'une des grandes transformations économiques à venir.
Non pas vendre une intelligence supérieure à Homo sapiens.
Mais vendre les moyens technologiques permettant à Homo sapiens d'apprendre davantage.
Reste alors une question fondamentale.
Si l'intelligence artificielle devient progressivement l'intermédiaire entre l'être humain et l'acquisition de ses connaissances, qui contrôlera demain l'accès à cet apprentissage : les individus, les institutions éducatives ou les entreprises qui possèdent ces systèmes ?
Car la véritable révolution ne sera pas seulement technologique.
Elle pourrait aussi redéfinir l'économie mondiale de la transmission des connaissances.
#IntelligenceArtificielle #Éducation #Apprentissage
Why can thirty minutes inside a video game sometimes feel like five?
The clock has not changed.
But attention, activity, novelty and memory influence subjective duration.
GTA VI offers another environment for exploring the difference between measured time and experienced time.
Knowing the map of GTA VI is not the same as knowing how to navigate it.
A player can memorize roads without anticipating traffic, obstacles or unexpected events.
Spatial memory provides information.
Intelligence uses that information to select actions.
Anthropic entre en Bourse avant OpenAI. Mais qui a réellement remporté la course à l'intelligence ?
La question mérite d'être posée autrement.
Imaginez Albert Einstein devant les valorisations financières de nos laboratoires d'intelligence artificielle.
Il pourrait examiner les milliards investis, les performances des modèles et les infrastructures nécessaires à leur fonctionnement.
Mais aucun de ces chiffres ne suffirait à établir une théorie scientifique de l'intelligence.
Car reproduire un résultat n'est pas nécessairement expliquer le mécanisme qui le produit.
Un ordinateur peut trouver une combinaison gagnante aux échecs. Cela ne signifie pas que son fonctionnement reproduit celui de la cognition humaine.
De même, un modèle linguistique peut produire un raisonnement convaincant sans démontrer que son architecture fonctionne comme celle d'un cerveau biologique.
Anthropic a franchi une étape financière historique.
OpenAI pourrait suivre.
Mais la compétition scientifique reste ouverte.
L'humanité saura-t-elle distinguer la valeur économique d'un système de sa capacité réelle à nous faire comprendre l'intelligence ?
Two players enter the same street in GTA VI.
One notices traffic. Another notices an escape route.
The visual information may be similar, but the selected information differs.
Human cognition does not simply receive an environment.
It acts according to what matters to the observer.