Victor Chapela retweeted
Great piece by my partner Konstantine, on how AI is a cognition revolution and how it compares to what we saw with the industrial revolution.
"Over the course of about two centuries, physical work went from 99% biological to 99.9% machine. [...] A century ago, 99% of cognitive work was done by humans. In the near future, 99.9% will be done by machines.
What stays economically human the longest is what was never really cognition to begin with: wanting things, choosing between them, being accountable for the choice, and being trusted by other people."
Agree
I’d argue the opposite: games will be one of the only categories of software that have value post-singularity.
Games are an expression of human cleverness and storytelling. Software is just the medium.
Today’s AI can make art better than some humans. As a result, utilitarian art is losing value. You might not pay someone to sketch you some logo ideas anymore; ChatGPT will do it for ~free. But nobody’s filling galleries with AI art, because the value of a painting isn’t the painting itself. It’s a person communicating ideas and evoking feelings. The painting is just the medium.
Utilitarian software is losing value too. ChatGPT will do it for ~free. Most software is about utility. Games aren’t.
1
184
Codex 5.6 Sol significantly superior than Fable 5
In continuous 5-day goal-based complex work with both Codex and Fable working side by side on the same problem:
- Codex with Sol is stable, resourceful, and has had only one drift case where I had to intervene.
- Codex with Sol follows instructions and change course with one single prompt.
- Codex with Sol has worked uninterrupted for > 5 days while I have been steering it successfully.
- Fable consistently forgets, drifts, and ends prematurely. Very hard to steer. Had a case where we logged 7 failed steering attempts.
- Fable when compacting memory completely wipes out relevant instructions which has lead to stopping shortly after context compact.
- Fable has required goal instructions to add backup heartbeat timers and reread instructions as part of its goal to keep it on track. Even with all of that, it fails.
- Fable drifts constantly off course and sometimes into deep rabit holes from which it never comes back out (even after acknowledging them).
- Fable on top of all of that eats up tokens faster and costs much more.
- Also tried Opus 5 for a while when it was released, problems are worse than with Fable.
1
93
Indeed.
Dimensionality reduction can see structures that do not exist and miss structures that exist.
𝐓𝐡𝐞 𝐬𝐢𝐦𝐩𝐥𝐞𝐬𝐭 𝐞𝐱𝐩𝐥𝐚𝐧𝐚𝐭𝐢𝐨𝐧 𝐢𝐬𝐧’𝐭 𝐚𝐥𝐰𝐚𝐲𝐬 𝐭𝐡𝐞 𝐛𝐞𝐬𝐭 𝐨𝐧𝐞.
Sometimes, no simple way of describing the complexity of the data exists, at least not one that humans can readily interpret.
Then, faithfully representing the data is at odds with finding simple explanations.
“Everything should be made as simple as possible, but no simpler”
A basic intro video with some strategies how to avoid mistakes when using dimensionality reduction:
youtu.be/nIB-WjWzDSc?si=vfKT…
4
1,413
extremely unprofessional. if kimi wants to make it as a frontier lab, they need to act like one: perhaps silently route people to worse models, and maybe write a blog post about the collapse of humanity
Kimi K3 has received far more love than we expected, and our GPUs are feeling it.
Over the past 48 hours, demand has pushed close to the limits of our current capacity. To protect the experience of existing subscribers, we're temporarily pausing new subscriptions and prioritizing compute for current members. Existing subscribed users are not affected.
We're adding capacity as fast as we can and will reopen new subscription spots in batches.
Going forward, we'll also split membership into two more focused plans: Kimi Membership for Kimi Web, App, and Work; and Kimi Code Membership for coding workflows. This will help us match compute more precisely and keep the experience stable.
Thank you for your patience and understanding!
Agreed. On top of that, many equate “scientific method” with “peer review academic publishing”. This is fundamentally broken in the new era of AI. It requires a set of average “human brains” that are supposed to not use “AI” because of academic ethics to check each others assumptions and results.
In the new era of complex dynamic systems and emergent properties, human brains are fundamentally limited. AI will find multidimensional non-linear relationships that are no longer fully understood by humans. This AI generated new “science” may never get through the human-referee-driven publishing bureaucracy. We will end up with new discoveries going directly from discovery → products or services - completely bypassing the human mind bottleneck.
Replying to @FutureJurvetson
There is no such thing as a scientific “method” in the sense of a reliable algorithm or repeatable process.
We’re taught in grade school that a scientist tabulates observations, then induces a theory from those observations, and then tests them to get to the truth.
But observations are themselves mini-theories (you never observe anything directly), induction from observations can’t give you novel theories (black swan problem), and testing can only eliminate competing theories, not prove them (error correction).
All you’re left with is creative guesswork and then eliminating the bad guesses to get you closer to the truth, but with no finality. Not much of a process or method.
“Scientists” have no special method or process that they employ. They’re doing what all of us do all day long to navigate the world, hopefully with more rigor and dedication, but it’s the exact same process at its core.
What made the scientific revolution possible was a philosophical shift in which parts of western society created social conventions around rigorous truth seeking rather than continuing to justify the beliefs of the ancients.
See Deutsch’s “The Beginning of Infinity” and Karl Popper’s writing, “On the Non-Existence of Scientific Method."
102
Many wonder why Silicon Valley engineers sometimes earn 10x or 100x more. They have learned how to scale to 100s of millions or billions of users. It is a different way of thinking and problem solving that would be an overkill in smaller scale projects.
Push-based systems come up in 90% of system design interviews.
Here's the exercise you should be able to solve:
Design a notification system for 100M users. Some have 50 followers. Some have 10M.
The instinct is to hold a WebSocket connection open to every active user and push updates as they arrive. Clean mental model. It collapses the moment a celebrity posts.
When someone with 10M followers posts, you push to 10M open connections simultaneously. Your message broker saturates. Your WebSocket servers fall over. The system fails at the exact moment it needs to work.
That's the fan-out problem. And it kills more interview answers than any other mistake.
The production answer: push and pull aren't binary. You pick based on follower count. Users with fewer than 1,000 followers get push fan-out. Each follower gets notified immediately.
Users with millions of followers get pull fan-out. Their feed assembles on read. Nobody gets a push. Followers see the post when they open the app.
Twitter built exactly this: push-on-write for small accounts, pull-on-read for large ones.
But fan-out is only half the problem.
Push means stateful connections. Your servers now need to know which connection lives on which machine. You can't route blindly. Most teams reach for Redis pub/sub here; the WebSocket server subscribes, the backend publishes, the message finds the right node.
Add a 3-second network drop and you have another layer: what did the client miss? Now you need sequence IDs, a message buffer, and reconnect logic that replays missed events.
"Push-based" became push with a pull fallback, a message broker, sticky routing, and a replay buffer.
Most engineers stop at the first diagram.
The ones who get the offer keep pulling the thread until the system breaks.
83
Reductionist thinking is limited in the type of problems it can solve.
All of our recent AI advances are emergent properties within dynamic complex systems. They cannot be understood or steered through reductionism.
The near future is all about steering complexity.
1
54
Intelligence can be defined as iterative energy efficiency optimization.
There are only two honest metrics when it comes to benchmarking intelligence: novelty and efficiency.
You don't need intelligence to solve a known problem (only memory). And you don't need intelligence to solve a problem via brute force. But to solve a novel problem efficiently, intelligence is the only way.
44
Good example that prediction can be disproportionately influenced by unlikeliness.
Polymarket prices are highly accurate in predicting future events. The source of that accuracy is less obvious.
In a new working paper, we find it is not the “wisdom of crowds,” but a small minority of informed traders.
Fewer than 3% of accounts appear to drive price discovery; most perform no better than chance.
The majority generates most of the volume but little of the information, effectively funding the informed minority.
Check the paper here: papers.ssrn.com/sol3/papers.…
51
Yes, totally agree with the concept of relational economy expanding.
New essay on the economics of structural change and the post-commodity future of work.
1. Almost any question about the impact of advanced AI on the economy needs to start at the same place: what is still scarce? Answer that, and the analysis becomes pretty straightforward. This essay explores what becomes scarce if AI really can replicate most of what humans do in production, and what this mean for the future of jobs.
2. My conjecture, working through the economics: labor reallocates across sectors, and the sector it reallocates to has properties that keep labor a meaningful share of the economy. Ultimately this is about the structure of demand itself. For this, we have to go back to Girard, Augustine and Rousseau: once people's base needs are met, their preferences shift to comparative motives (e.g., status, exclusivity, social desirability). This motive is inherently non-satiated.
4. The key paper is Comin, Lashkari, and Mestieri (Econometrica 2021). As people get richer, they don't buy proportionally more of everything. They shift spending toward sectors with higher income elasticity. They estimate income effects account for 75%+ of observed structural change.
5. The ironic consequence: the sector that gets automated becomes a smaller share of the economy, not a larger one. Agriculture got massively more productive and its share of employment collapsed. Manufacturing too. The "stagnant" sectors absorb the spending and the jobs.
6. So the question is: which sectors have high income elasticity in a post-AGI world? I argue it's what I call the relational sector. Categories where the human isn't just an input into production, it is part of the value.
7. Why does the relational sector have high income elasticity? Because human desire has a mimetic, relational dimension. We don't just want things for their intrinsic properties. We want what others want, and we want it more when others can't have it. Girard, Rousseau, Augustine, and Hobbes all saw this.
8. In work with Kristóf Madarász, we showed this experimentally: WTP roughly doubles when a random subset of others is excluded from the good. And in new work with Graelin Mandel, AI involvement kills the premium. Human-made art gains 44% from exclusivity; AI-made art only 21%.
9. This all comes together for the core argument. The sector that absorbs spending as AI makes commodity production cheap is one where human provenance is part of the value, and demand for it grows faster than income. Exactly the profile that keeps labor meaningful.
10. To be clear about the claim: I'm NOT saying aggregate labor share must rise. It may fall. The claim is about sectoral composition, i.e., where expenditure and employment go once commodities get cheap, and the fact that the sector that will absorb reallocated labor maps to a substantial component of human preferences and desire.
11. If you're interested in the formal model, a linked companion technical note works out all the economics.
Read the essay here: aleximas.substack.com/p/what…
1
1
47
This.
Replying to @eladgil
11/12 AI is about selling work, not software. Units of labor as the product
This is a big shift in TAM, agents, how to build product in the AI era. Many AI markets are 10-100X the size of their seat based software counterparts
32
True
As Carl Jung put it, "Man needs difficulties. They are necessary for health." Yet most people instinctually avoid pain. This is true whether we are talking about building the body (e.g., weight lifting) or the mind (e.g., frustration, mental struggle, embarrassment, shame)--and especially true when people confront the harsh reality of their own imperfections. #principleoftheday
25
































