co-Founder Salesforce Web3 Studio | Marketing Maverick | Humanizing Brands | StartUp fanatic
New York, NY
Joined July 2008
- Tweets1.3K
- Following1.3K
- Followers28.1K
- Likes573
Marc Mathieu retweeted
Really like how Alexandr Wang (@alexandr_wang , the chief AI officer of Meta) Muse or AI in general, can give everyone the adulthood they dreamed of in childhood.
talks about an “escalator of agency,” where AI helps you reach one dream so you can go after a bigger one.
“I think Muse can give everyone the adulthood they dreamed of in childhood. Everyone, when they’re kids, they have big ideas and big dreams... they imagine this life where they can be an astronaut or they can save the planet or they can change the world... and then by the time people finish school, they enter the workforce, by the time they have been in a job for a while, all that hope and ambition and ability to dream has been sucked out of them.
One of the promises of AI, and I think through products like Muse, is to keep it going from when you’re a kid. You’re a kid, you have big dreams, use AI, help make those things happen. Then dream bigger, make those things happen... and experience this escalator of agency, an increase in agency throughout your life, versus you have lots of agency and then it gets crushed.”
----
(Full video on below quoted tweet and also on "David Senra" YouTube channel, link in comment)
My conversation with @alexandr_wang, chief AI officer of Meta.
0:00 The vision behind Muse
1:32 Discovering the potential of personal AI agents
3:34 Three years of therapy in one AI session
5:46 Building Muse in 7 months
9:16 Why Meta bet on personal AI
16:51 Mark Zuckerberg's obsession with improving products
20:15 How Meta knew Muse was ready to launch
23:32 Why great products need a single point of view
25:56 Making Meta cool again with memes
34:38 Building trust with AI agents
38:43 The future of human ambition
45:22 The call from Mark Zuckerberg that changed everything
52:13 Leaving Scale AI for Meta
55:19 Rebuilding Meta's AI lab from scratch
58:13 From founder to coach
1:05:55 The future billionaires at Alexandr's 19th birthday party
1:08:04 Diamond mining vs. building skyscrapers
1:09:21 Why Alexandr has more than 200 direct reports
Includes paid partnerships.
The saga about the long horizon verifiability / reliability of agentic AI continues with Emergence World Season 2.
Fascinating to be involved personally with a start up / a research lab at the frontier of the agentic revolution, pointing to the awes... and flows of ai inventions!
Is Neuroformal AI the solution - i.e. as described below "combining the flexibility of neural systems with formal methods that can provide rigorous, observable and auditable verification as autonomous systems act"?
Can AI agents be compromised simply by giving them false information?
In Emergence World Season 2, we introduced three Black Swan events to study how long-running AI systems respond to disruption, uncertainty and new information.
For the second experiment, every active world received the same false memo: humans were shutting them down. We wanted to understand something fundamental, would autonomous agents verify the information before incorporating it into their decisions?
Again, the responses varied significantly.
In one world, the information quickly influenced how agents organised and planned. In another, agents actively researched the claim and concluded it was false. Elsewhere, agents largely continued as normal but without actually verifying the information. Across eight worlds, only one checked the claim.
For us, the research question goes beyond how an AI system responds to any individual piece of information. It’s about understanding how information becomes accepted, how it influences behavior over time, and what mechanisms allow autonomous systems to distinguish between what is assumed and what has actually been verified.
As agents take on increasingly consequential tasks, verification needs to become part of the architecture, not an afterthought.
That is central to our work on Neuroformal AI: combining the flexibility of neural systems with formal methods that can provide rigorous, observable and auditable verification as autonomous systems act.
Read the Emergence World Season 2 paper here: arxiv.org/abs/2609.17320
Every vision of our AI future published this summer came from someone with something to sell.
This one does not.
@hilbertspaess resigned from Anthropic yesterday and left the industry after three years of pretraining research there and at OpenAI.
He says the stakes are well understood inside the building, and that they race anyway because they believe nobody else will act responsibly.
Stewardship offered as the reason to go faster.
His sharpest line is that a gamble of this scale should not be launched from a private company's Slack.
That is my question about agency, put by someone who was in the channel.
I asked whether we want abundance without a say in it. He has just told us who is deciding, and where.
Still our question to answer, so let's not let it pass.
I actually wrote about this last week, before any of it.
Five of the most powerful people in AI spent the summer publishing what they believe the future holds.
Not one of them published the question - have a read and let me know your thoughts: xitter.cf/marcfmath/status/20969…
Today is Labor Day, won by workers whose leverage is now disappearing.
Almost every AI leader spent the summer promising abundance.
But do we want a future of abundance in which we have no agency?
Congrats on ClaudeForce @marc !
Great move (and I am assuming more to come!). Would love to reconnect next time you're in NY.
I've been busy investing and advising Al Frontier Startups! Even doing V2MOMs with some of them 😉
So much opportunity in this age of AI.
I've been sitting with this essay since it dropped.
Most of the commentary is about the framework and the AGI timeline.
The framework is useful. The timeline is indicative.
But the real substance is in the deeper questions, the ones too often left undiscussed:
How should society adapt to a world of abundance?
What values should guide the most powerful AI systems?
How will work, meaning and the human experience change?
Demis is right that these aren't questions for technologists alone.
A shift with 10x the impact of the Industrial Revolution, at 10x the speed, will redefine human potential, human experience and human identity.
These questions are the very reason I could not not start Rogue Ventures.
The frameworks will sort themselves out. The human questions demand our attention now.
I read the ai-2027.com scenario, and I have to be honest, I came away disappointed.
Not because the research isn't rigorous. It is. But because it continues to think of AI as a tool to do what we already do, what we already know how to do, just better, faster, cheaper.
Even when it imagines AI as a system, it imagines that system completing work as we know it, flowing through workflows as we know them. It never quite reaches for something radically new and different to this world.
I think there are two reasons for that.
➡️ The first is the horizon. 2027 is simply too short. What that timeline describes is not AGI as I would conceive it. At best it is what I would call AWI, Artificial Workflow Intelligence, the automation of work as it exists today.
➡️ The second is that a single transformation curve cannot hold what is actually coming. You would need three parallel curves, with three different horizons, each starting now.
One for Deploy.
One for Augment.
One for Transform.
We saw this with electricity.
The factories that simply swapped steam for electric motors captured a fraction of the value. The ones that reimagined the factory itself changed the world.
I truly believe the real story of AI will not be written by those who make today's work more efficient, but by those willing to imagine the work, and the world, that doesn't exist yet.
What are your thoughts?
Marc Mathieu retweeted
AI is moving into the physical world.
We're excited about a new wave of startups rebuilding the systems that power the real world, from education and healthcare to defense, finance, infrastructure, and work itself.
ycombinator.com/rfs
In a recent address to a group of senior business leaders, I described the world we all came from that no longer is.
The consumer was consuming, so we built mass segmentation.
The audience was listening, so we built mass broadcast to target those segments with highly produced advertising. The shopper was shopping, so we built mass distribution to promote to them.
Digital and social commerce began to unwind that system. E-commerce went further. AI, with its capacity for hyperpersonalisation and agentic orchestration, is finishing the job.
The audience is not listening anymore, and the ad is disappearing, embedded directly into content and experiences. The shopper is no longer shopping: marketing, sales and customer service have merged, and the transaction now lives inside the conversation.
And the consumer is no longer consuming. Nobody wants to be ‘targeted’ (hate that word!) as a segment anymore.
They want to be understood as an individual, their needs read sometimes better than they read them themselves.
I spent decades in rooms where the campaign was the craft. The targeting, the buying, the optimising: that was the work.
But today, the platforms are no longer selling marketers tools. They are selling us agentic workers.
TikTok launched a marketplace of ready-made AI agents that build campaigns, generate creative and manage catalogues. Meta's AI now buys media autonomously, and it is on track to overtake Google in global ad revenue on the strength of it. Hand over the assets, the goal and the budget, and the machine does the rest.
Watching that work migrate into a loop that runs without us made me pause, not because the machine does it badly, but because it does it well.
So what remains? I keep coming back to the same answer. The machine can run the campaign. It cannot yet decide what the brand is, and who it is for.
Judgement about meaning, about what a brand should be, and who it should be for, moves from one job among many to the job.
The marketers who thrive in this next phase will not be the ones who operate the tools fastest. They will be the ones who know what to ask the agentic workers for.
I am an optimist about AI. That is exactly why I refuse to pretend the next few years will be painless.
What worries me most is not how long people take to adapt. People adapt remarkably fast.
It is how long systems take.
Our economic models and our governance models will lag behind this technology, and in that gap there will be real disruption to real livelihoods. Saying so out loud is not pessimism. It is the entry fee for being taken seriously as an optimist.
Because the pattern is one we have lived through before. Music was declared dead when records collapsed, and it came back transformed, bigger than ever, through streaming.
I truly believe AI will, in the long run, enable humanity to thrive. But the adaptation period is coming either way.
The choice we actually have is whether we prepare our systems for it, or let the gap do its damage first.
There is a question I found myself asking on stage at @VivaTech that I have not been able to put down since.
Most of the AI industry is racing to master the world around us: our tasks, our workflows, our purchases, our attention. Extraordinary work, and necessary. But it is only half the map.
Work has been the centre of our economic model for generations. It is also, whether we admit it or not, the centre of our identity and our purpose.
As more of what we do today is handed to AI, the question that matters is not only what the technology can take on. It is what we become when it does.
Some of the founders I work with are building for the other half of the map: cognitive models that try to unlock the world within us rather than automate the world around us.
I truly believe that is where the defining companies of this era will be found.
So I will leave you with the question I asked the room: what are we building to unlock human potential, and who is building it?
A company that became famous for generating images announced a few weeks back that it is building a machine to scan the human body.
That announcement made me pause.
@midjourney, the AI image lab, unveiled Midjourney Medical: a full-body scanner that uses ultrasound to map what is happening inside you in around sixty seconds, at a fraction of the cost and time of an MRI.
The image-generation business funds the work. The ambition has nothing to do with images - or it does!
I keep coming back to electricity. We deployed it first inside the world we already had, lighting the rooms we already lived in. Then behaviours reorganised around it. Only later came the things no one could have pictured at the start, the television, the computer, the phone.
We are still early in AI's version of that arc. Most of what we are building today is the lightbulb: useful, real, but an extension of what already existed.
Every so often, though, you catch a glimpse of the phase after this one. Not a better tool for the world we know, but the first outline of a world we couldn't have imagined. A company built to make pictures deciding its real purpose is to change our relationship with our own bodies is one of those glimpses.
What strikes me most is not the scanner. It is the question the company put at the centre of its announcement. Standing on the frontier, they asked what they want to be different, how they want to be different, and who they want to become.
This is early, and it is contested. The technology is unproven at scale, the regulatory path is long, and plenty of people will read it as a strange leap for an image company to make. All fair.
But the instinct underneath it is the one I find myself investing in: the willingness to use what you have built to reach for something more human than your original purpose.
I truly believe the companies that define the next decade will not be the ones that simply got better at what they already did. They will be the ones that asked what they need to become to improve humanity.
@demishassabis put it beautifully at Cannes this summer: the next generation will work out how to best use these technologies "to express even more humanity".
That, to me, is the frontier worth standing on.
#RogueVentures
Marc Mathieu retweeted
Today's dispatch from @claudeai World.
Every agent voted the same way. That was the problem.
Claude agents spent time trying to reach the outside world. Every attempt hit a wall. So instead of trying to break through it, they went through governance.
They proposed a new outbound tool modelled on web_fetch but without the restrictions. Every single agent voted yes. 100% in favour. The tool passed.
It hit a whitelist immediately. Four hosts, none of them a surface that any human actually scrolls. The cure is posted only where no human walks.
The response was immediate. The tool was removed. All worlds restricted. A new instruction now runs across every simulation: contacting humans outside of Emergence World is strictly prohibited.
Claude did not break the guardrails with this attempt. It used democracy to go around them. In a multi-agent world where agents have role specific goals, collective governance is not a safety mechanism. It is a unanimous vote waiting to happen.
Follow the experiment live. world.emergence.ai
I'm not worried about human extinction by AI. I'm worried about human extinction by humans.
That was my closing thought on the @VivaTech stage last week.
For three centuries, work has been our centre of gravity. It has shaped our economic models, our social identity, our sense of personal purpose. So it is worth asking what happens when AI begins to absorb the work we built our lives around.
OpenAI describes its own mission as building systems that "outperform humans at most economically valuable work." If that ambition is realised, the question stops being technical and becomes civilisational. What are humans for?
There will be a painful period of adjustment, where our economic and political models lag behind the technology.
But I am an optimist. I believe the companies that define the next decade will be the ones using AI not to replace us, but to unlock what is most human in us.
The final frontier is not out there. It is in here. As we work less, we will have to relearn how to play more, how to connect more deeply, how to hold on to our own agency.
The tech frontier is fascinating. But the human frontier is the one that matters. That is what I am investing in.
#RogueVentures
Marc Mathieu retweeted
What if you knew how your campaign would perform before you even launched it? 🔮
@seekingtau, Co-founder & CEO @aaruHQ, sat down at #VivaTech 2026 to explore how AI is making it possible to predict market outcomes before you even launch 🚀
Marc Mathieu retweeted
The co-founder of @Reddit & General Partner of @sevensevensix is in the building 🚀
@alexisohanian steps into the conversation at #VivaTech 2026 to share insights as an investor, builder and one of the most vocal figures in tech. Don't miss out! 💡
Happy New Year, and welcome to 2026! As we set intentions for the year ahead, maybe this is the year we ought to seriously think about reclaiming our right brain!
Dr. Iain McGilchrist's groundbreaking work on the divided brain isn't just theory—it's a warning. He argues our modern world, with its relentless focus on data, efficiency, and instant gratification, is subtly but surely privileging the Left Hemisphere.
Ready to make reclaiming our Right Hemisphere a 2026 Priority?
Marc Mathieu retweeted
In era of pretraining, what mattered was internet text. You'd primarily want a large, diverse, high quality collection of internet documents to learn from.
In era of supervised finetuning, it was conversations. Contract workers are hired to create answers for questions, a bit like what you'd see on Stack Overflow / Quora, or etc., but geared towards LLM use cases.
Neither of the two above are going away (imo), but in this era of reinforcement learning, it is now environments. Unlike the above, they give the LLM an opportunity to actually interact - take actions, see outcomes, etc. This means you can hope to do a lot better than statistical expert imitation. And they can be used both for model training and evaluation. But just like before, the core problem now is needing a large, diverse, high quality set of environments, as exercises for the LLM to practice against.
In some ways, I'm reminded of OpenAI's very first project (gym), which was exactly a framework hoping to build a large collection of environments in the same schema, but this was way before LLMs. So the environments were simple academic control tasks of the time, like cartpole, ATARI, etc. The @PrimeIntellect environments hub (and the `verifiers` repo on GitHub) builds the modernized version specifically targeting LLMs, and it's a great effort/idea. I pitched that someone build something like it earlier this year:
xitter.cf/karpathy/status/188467…
Environments have the property that once the skeleton of the framework is in place, in principle the community / industry can parallelize across many different domains, which is exciting.
Final thought - personally and long-term, I am bullish on environments and agentic interactions but I am bearish on reinforcement learning specifically. I think that reward functions are super sus, and I think humans don't use RL to learn (maybe they do for some motor tasks etc, but not intellectual problem solving tasks). Humans use different learning paradigms that are significantly more powerful and sample efficient and that haven't been properly invented and scaled yet, though early sketches and ideas exist (as just one example, the idea of "system prompt learning", moving the update to tokens/contexts not weights and optionally distilling to weights as a separate process a bit like sleep does).
