My isometric map of Toronto now covers the whole City: all 630 km², from Etobicoke to Scarborough and from the lake up to Steeles. And all of it is redrawn in a new, crisper style. 🧵👇
@lemire@ChristianJauvin A philosophical view on problem solving with AI: at what level of abstraction are students obligated to understand the AI solution?
It's like advising a brilliant PhD student. I gotta keep up the relevant literature review, theory and technology.
Most people will happily play stupid games if it gets them prestige (or prestigious mates). But, ultimately, society only progresses if people have actual skills.
I am working with my colleague @ChristianJauvin on a short program on AI-assisted programming for Quebec software developers who want to upgrade their skills. Evidently, students will have to solve problems that they could not solve without AI.
Later in October, I am proud to be giving the Sedgwick Lecture to a bunch of undergraduate students. I will be giving them good and bad news. The bad news is that it is all over. Much of what they were taught must be rethought. The good news is that they may soon start their careers at a historical time.
Expect me to write a lot more on AI-assisted programming.
I believe we need to make a deliberate effort to keep humans in the loop in all critical processes across our economy and society, regardless of whether it is technically necessary.
Even if AI develops the *capability* for advanced autonomy, we should not make it highly autonomous. We have to maintain control and keep visibility and understanding of all critical processes, we should not blindly hand over everything to AI agents just because we can. AI as a tool in the human hand is the only form of AI that is worth pursuing.
I can’t say enough good things about John Carmack @ID_AA_Carmack and his Keen Technologies. But now Khurram Javed @kjaved_ and I have broken away to start our own startup and pursue a slightly different path toward understanding intelligence. Like Keen (and like Ineffable) we at Oak Lab @oaklab believe in reinforcement learning and that intelligence is created and maintained from run-time experience. But we think current deep learning methods are weak and inefficient, and need not more tweaks, but fundamentally new ideas and a thorough reworking before they can provide a solid foundation for achieving the more ambitious goals of AI.
If Steve Jobs were still alive, he would have the moral authority to face and maybe even to solve this problem. But I doubt anyone in the phone business now does.
@lemire I meant the handwritten ones. There is just a particularly gossipy quality that comes through human written content, found only between the lines.
Wow I can already say after just 5 hours using @AnthropicAI Opus 4.7 that this is the first model that "gets" what I'm doing when I'm working. It feels aligned with me in a way no previous model did.
(4.6 actively worked against me. I hated it. So this is *very* exciting!)
A few years ago, a small revolution occurred. Two major companies, Coursera and Udemy, began offering online courses. They made university-level content cheap and widely available.
For a modest fee, you can now take classes on almost any topic and even earn a certificate to add to your resume.
Some claimed this would disrupt higher education.
It did not.
The fundamental reason is a misunderstanding of what education really is.
Consider this: high school students across the USA are currently choosing colleges. How many are investigating the quality of the classes or measuring how much students actually learn?
Very few. Students care about prestigious names, strong sports teams, and social life. Class quality is rarely a factor.
Even at top schools like Harvard, attendance is often low. Professors are hired and promoted based on research, not teaching ability.
If you offered young people four years at a prestigious university for free—on the condition that they could never mention they attended it—few would accept. Yet if education were truly about learning, they should.
Improving the teaching itself changed little. It is not why most people attend university.
However, the online revolution did happen. In 2021, the vast majority of university students had taken at least one online class. The numbers have dropped slightly, but nearly 50% still do today.
Are these online courses any good? We do not really know—because we also lack clear measurements of how well traditional classroom courses perform.
Today, anyone can access the best lectures, the best books, and a highly personalized tutor (just ask an AI).
The scenario in Good Will Hunting, where an unschooled young man outperforms Harvard students, feels more credible than ever. You could always buy or borrow books. Now you can do far more.
Does this mean no disruption in higher education? No. There is a slow, ongoing disruption. Year after year, elite universities are slightly less revered, while alternative institutions gain respect. New elites emerge and some decline. But the process is very slow, unfolding over decades. It is too slow of a tech startup.
The linear model of innovation is a theory of technological progress. It is also ridiculously wrong.
The modern version claims: the government funds professors who do wild research; ideas then flow into industry and become customer products.
Who defends this model at every opportunity? Professors who receive the funding, of course.
I am not a libertarian, but I dislike misleading, self-serving, and childish models.
Consider the claim that government-funded research led to GPU computing. It implies that without government grants to professors 30 years ago, we would not have Nvidia cards running AI models today.
The demonstration is simple: professors apply for funding, hire grad students, and those students later join companies like Nvidia. Voilà—causation established.
This sounds plausible and suggests that more money to professors means more innovation, while zero government funding means zero innovation.
But we already ran that experiment. Before the 1960s, the U.S. government rarely funded professors. It did occasionally—for example, it funded Professor Langley’s failed attempt to build an airplane. His crashes led the New York Times to declare that humans would never fly. Meanwhile, two brothers with no government funding succeeded.
We got the laser, the transistor, nuclear technology, and more. Contrary to what you may have heard, Einstein did not begin his career by writing research grants.
By many measures, we have seen relative stagnation since the 1970s, when large-scale government funding of professors began. Outside computing, progress has been slow. We declared war on cancer in the 1970s and still rely mainly on poison (chemo), cut (surgery), or burn (radiation).
Some innovations do emerge from universities, but the theory that innovation primarily comes from government grants to professors is ridiculous. It takes a PhD to believe such nonsense.
“If you’re so smart, Daniel, where does innovation come from?”
Innovation is disruption, not a routine output of a machine. It is a complex process.
A more useful model: innovation is often customer-driven. Where customers have imagination and money to spend, innovation follows.
We got the Industrial Revolution because women wanted nice underwear. We got the piano because people wanted beautiful music. We got powerful GPUs because boys and men wanted to play action video games and were willing to pay for them.
Cheap, powerful GPUs became widespread, and many young engineers learned to program them because they loved games.
Much of today’s AI work is driven by advertising: companies want to reach customers, so they run vast numbers of GPUs to sell more stuff.
A better model of innovation is this: customers want things, industry works to provide them, and eventually the resulting technology reaches university professors, who incorporate it into their courses and research.
We see it around us. My wife wants to consult ChatGPT about dinner. The people running ChatGPT improve it so she will pay or accept ads. Only later does the technology get studied on campus.
And we know that's more true that the ridiculous linear model because we don't see the OpenAI engineer spending all their time on campuses to find out what to build next.
How can customer drive innovation? By providing the forcing function of natural selection.
In my doctorate, I proved the Erdős Primitive Set Conjecture, showing that the primes themselves are maximal among all primitive sets.
This problem will always be in my heart: I worked on it for 4 years (even when my mentors recommended against it!) and loved every minute of it.
[Primitive sets are a vast generalization of the prime numbers: A set S is called primitive if no number in S divides another.]
Now Erdős#1196 is an asymptotic version of Erdős' conjecture, for primitive sets of "large" numbers.
It was posed in 1966 by the Hungarian legends Paul Erdős, András Sárközy, and Endre Szemerédi.
I'd been working on it for many years, and consulted/badgered many experts about it, including my mentors Carl Pomerance and James Maynard.
The the proof produced by GPT5.4 Pro was quite surprising, since it rejected the "gambit" that was implicit in all works on the subject since Erdős' original 1935 paper. The idea to pass from analysis to probability was so natural & tempting from a human-conceptual point of view, that it obscured a technical possibility to retain (efficient, yet counter-intuitve) analytic terminology throughout, by use of the von Mangoldt function \Lambda(n).
The closest analogy I would give would be that the main openings in chess were well-studied, but AI discovers a new opening line that had been overlooked based on human aesthetics and convention.
In fact, the von Mangoldt function itself is celebrated for it's connection to primes and the Riemann zeta function--but its piecewise definition appears to be odd and unmotivated to students seeing it for the first time. By the same token, in Erdős#1196, the von Mangoldt weights seem odd and unmotivated but turn out to cleverly encode a fundamental identity \sum_{q|n}\Lambda(q) = \log n, which is equivalent to unique factorization of n into primes. This is the exact trick that breaks the analytic issues arising in the "usual opening".
Moreover, Terry Tao has long suspected that the applications of probability to number theory are unnecessarily complicated and this "trick" might actually clarify the general theory, which would have a broader impact than solving a single conjecture.
Computer science is gradually returning to the domain of physicists, mathematicians, and electrical engineers as large language models automate much of what we currently call software engineering.
The field’s center of gravity is shifting away from manual code writing and toward deeper theoretical thinking, mathematical insight, and systems-level reasoning.