This is one of the most underrated posts of 2026.
Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel.
- Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all.
- Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract.
- Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code.
- And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while.
The funnel is driven by a combination of factors:
1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel.
2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel.
3. Access. Free-tier, paid-tier, internal.
So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.
Taking a moment to appreciate how far we have come to enable the perfect snipe. All this in 1/10 of a second 😎
Good perspective for AI. We are still early 🤩
youtube.com/shorts/4OJteWc38…
Danial Jameel retweeted
Banking operations are where AI gets tested for real.
Saris raised a $28.8M Series A led by @8vc, with @AudaciousHQ and @homebrew joining, as @danialj, Alice Dinu, and James Dang build agentic workflows for lending, compliance, and operational tasks inside the systems banks already use.
The next banking tech winners will not rip out the foundation. They will automate the manual work hiding underneath it.
The Canadians that founded @ycombinator-backed startup Oohlala Mobile have a new venture they say will automate grunt work at banks and credit unions.
@danialj says his vision is "a future where humans and AI work side by side in financial services.
betakit.com/saris-raises-28-…
Danial Jameel retweeted
Try to build as much code as possible over the next few months. The prices you are seeing now for AI will probably not last too long.
🦔Microsoft canceled its internal Claude Code licenses this week after token-based billing made the cost untenable, even for a company with effectively infinite cloud resources. Uber's CTO sent an internal memo warning the company burned through its entire 2026 AI budget in just four months. American AI software prices have jumped 20% to 37%, and GitHub (owned by Microsoft) is dropping flat-rate plans for usage-based billing across its products.
My Take
The AI subsidy era is ending in real time. The same company that put $13 billion into OpenAI and built the Azure infrastructure powering most of Anthropic's compute just looked at the bill from a competitor's coding tool and decided it was not worth paying. That is not a productivity failure on Anthropic's end. Token-based pricing is forcing every enterprise customer to confront the actual cost of running these models at scale, and the number turns out to be far higher than the flat-rate experiments suggested.
This ties directly to my Gemini Flash post yesterday. Anthropic, OpenAI, and Google all raised effective prices in the last six months. Enterprises that built workflows assuming AI costs would keep falling are now watching annual budgets evaporate in months. Two outcomes look likely from here. Either enterprises scale back AI usage to fit budgets, which slows the revenue ramp the labs need to justify their valuations ahead of IPOs, or the labs cut prices and absorb the losses, which makes the unit economics worse at exactly the wrong moment. Both paths land in the same place, the numbers stop working, and somebody has to take the writedown.
Hedgie🤗
Liftoff.
The Artemis II mission launched from @NASAKennedy at 6:35pm ET (2235 UTC), propelling four astronauts on a journey around the Moon.
Artemis II will pave the way for future Moon landings, as well as the next giant leap — astronauts on Mars.
Danial Jameel retweeted
5 minutes ago, @karpathy just dropped karpathy/jobs!
he scraped every job in the US economy (342 occupations from BLS), scored each one's AI exposure 0-10 using an LLM, and visualized it as a treemap.
if your whole job happens on a screen you're cooked.
average score across all jobs is 5.3/10.
software devs: 8-9.
roofers: 0-1.
medical transcriptionists: 10/10 💀
karpathy.ai/jobs
Danial Jameel retweeted
After 16 years and 10 months -- I'm finally writing my first...X? Also, will start writing more regularly.
My first post is, ironically, on why I'm posting. qy.co/new
Danial Jameel retweeted
Interesting data from Wharton showing that 75% of companies see a positive ROI on AI investment.
But what’s more fascinating, and revealing, is the delta between companies that are < $2B in revenue vs. those that > $2B in revenue. Much higher rates of seeing significant gains and much lower rates of still being stuck in pilot phase for the smaller company cohort.
This is certainly intuitive as small companies can always move faster and adopt new technology quicker. But AI magnifies this classic difference because it changes the nature of work vs. just is another technology that you implement and move on from.
The amount of output you get from AI agents will be directly correlated to how much you change (or reset) your workflow. It’s a continuum from those designing a process from scratch at one end to those not willing to change a thing at the other.
The smaller companies definitionally can change their processes more easily or start from the ground up. This is clearly harder in big companies. But it’s often going to be necessary to fully get the gains from AI.
This will remain one of the big advantages that smaller companies and faster moving teams have for the time being. Will be interesting though as the larger companies begin to figure this out and catch up.
Danial Jameel retweeted
Our Co-Founding Partner @LaurenKolodny on stage at SF #TechWeek, leading a fireside chat on AI’s transformative role in financial operations.
Together with Erik Zhou (@brexHQ), @sashaorloff (@puzzlefin), and @danialj (Saris AI), they explored the potential of next-gen financial tools and how finance teams can power the growth of a company.
Danial Jameel retweeted
It is insane that just 18 months ago the state of the art AI model had a 4K token limit, and today state of the art can reach 1M tokens. There are essentially no examples in history of a technology improving at a rate of 250X in under 2 years.
Danial Jameel retweeted
Many of the biggest opportunities in AI will be taking functions or processes in business that are artificially constrained by their cost or complexity, and scaling them up to 100X more customers than were previously served. SaaS did this for software. AI will do this for labor.
Danial Jameel retweeted
In case you were wondering how superconductors work, this is a good explanation. They are a very interesting phenomenon and may prove economically useful, but are not a hard requirement for a sustainable energy future.
Physics has many powerful tools of reasoning to understand and predict reality. Those tools, like first principles analysis and thinking in the limit, are broadly applicable to anything, in my experience.
With Grok, @xAI is attempting to create an AI that reasons from first principles, which is fundamental if you care about getting as close to the truth as possible.
The acid test would be reaching a conclusion that is correct even if it is extremely unpopular, which means being right even when the training data is almost entirely wrong.
For example, Galileo concluded, after observing the moons of Jupiter from a telescope he engineered, that it was far more probable that Earth revolved around the Sun than the other way around. This view was so unpopular that he was forced to recant and placed under house arrest!
If you had trained an LLM on material back then, it would’ve given you the popular, but wrong, explanation. Due to social and legal pressure, it likely wouldn’t even acknowledge the possibility that the Earth revolved around the Sun.
For AI to help us understand the true nature of the universe, it must be able to discard the popular, but wrong, in favor of the unpopular, but right.
Replying to @pronounced_kyle
Superconductivity requires a spin-up electron to pair with a spin-down electron, so that the pair has a net spin of zero and therefore obeys Bose-Einstein statistics, not Fermi-Dirac. This breaks them free from the Pauli exclusion principle, so you can have a great many of them occupying a low energy quantum state that can slide effortlessly through the material like a ghost. Every superconductor relies on exactly that effect.
What differs between materials is the way in which the electrons pair. Electrons don’t like to pair, because of their mutual Coulomb repulsion, but in some materials they can both jiggle the crystal lattice in a way that weakly brings them together, like an arranged marriage for a couple that’s not really into each other. This is called phonon coupling, and can be modeled with BCS theory, at least… for simple crystal structures. That model predicts that superconductivity can only occur up to like 30 K, beyond which the lattice is jostling around too much because of temperature, so the pairing mechanism is disrupted.
However, there are some materials with complicated crystal structures, that get around the assumptions of BCS, and that are more robust when it comes to the effect of temperature on the pairing mechanism. For example, ReBCO, which I work with, can go up to ~90 K. This is a shockingly high number from the perspective of BCS, which is why these materials are called “high temperature” superconductors, despite still having to be pretty cold (liquid nitrogen boils at 77 K).
Anyway, the high T superconductors have more complicated and exotic pairing mechanisms. It’s no longer “simple” phonon coupling, but something that involves longer-range and more geometrically complicated details, and to this day it’s still not fully understood.
Although, the way supercurrent flows is well-understood, in all superconductors. It’s called the Ginsburg-Landau model, and is based on the idea that there’s some energy associated with electron pair formation, and some energy associated with pair-to-pair interactions, in addition to the usual electromagnetic terms. This model works extremely well, and is the basis for all superconducting technology.
One of the predictions of GL is that, when the condensation energy is nearly zero (around the Tc), there will be fluctuations where electrons temporarily pair and un-pair. These oscillations are extremely rapid, so in experiential measurements they will average out into a state that seems to be kinda superconducting, but not completely. But it’s really just fluctuating, a second-order phase transition. So that’s one cause of the transition width.
Another cause of the transition width is the purity of the sample, or lack thereof. A sample with varying quality throughout will have a wider transition, as some parts of it “turn on” before others.
With ReBCO, transition widths are typically a few K, with tails noticeable out to like a 10 K span, sometimes more. And dirty samples might have double or triple transitions, if different parts of it are oriented different ways.
Anyway, sorry for the novel of a response, but I hope that provides some context. Any Tc curve on a new superconducting sample is probably going to have quite a wide transition, for lack of sample purity, and also due to the fluctuations that thermodynamically should be there no matter what. So it seems extremely unlikely that a material which somehow superconducts at 4X the temperature of ReBCO would also have a much cleaner Tc transition.
Danial Jameel retweeted
Replying to @patrickc
People mock the overconfidence of startup founders, but there's a reason for it. Only people who are naturally somewhat overconfident will take on problems that, when viewed rationally, would seem too hard.
Danial Jameel retweeted
"We do these things not because they are easy, but because we thought they were going to be easy" is a surprisingly profound quote. When I ask people who've pulled off remarkable things, it's interesting how many confirm that they wouldn't have started if they'd know how long and difficult the road would be.
Danial Jameel retweeted
One day, when we have automated asteroid mining, this question will seem quaint.
By 2050 the world will need 1.4 billion tonnes of new copper to reach net zero. How will we find it?
Click the image to read the full article
Partner Content with @bhp

























