


Economics & AI @ @METR_Evals (ex-openai) tecunningham.github.io


Recently i've found myself getting more upset with discussions of AI risk on X Like, often it feels like the sceptics are just dunking on ppl who are worried And it's tough bc these issues are really complicated. I don't think there's any water tight case for an intelligence explosion happening soon, or it leading to AI takeover So there are loads of genuine uncertainties and ambiguities, and of course sceptics can and should raise them. But it's upsetting when they present them as slam-dunk proofs that the whole AI risk is flawed and silly. Like, do you really think that no one has thought of this before? Are you actually curious about whether your objection holds up? Remember how most issues are really complicated, and ppl are often wrong about stuff? And the bigger picture here is that hundreds of top experts inside and outside the industry think this will be a complete disaster. Like, that is obviously fucking scary and we should take it very seriously, not notice an apparent flaw in the argument and move on. To be clear, AI safety ppl can be annoying too. Historically, I've found them much more annoying to be honest. I've felt pressure not to say stuff that vibes as 'anti-safety', and noticed ppl engaging in bad faith with me when i do. I know none of this is specific to AI safety. Public discourse is dogshit on every topic. And probably X is waay better than average. But i still think it's a disgrace how low the standard is. For a topic this important everyone on both sides should try to do better

When Aristotle is reincarnated he will not say his problem was lack of data. He will see that we're working with mostly the same old jigsaw pieces but we've put them together in a different way, & he'll recognize it's better. The dolphin is next to the dog, not the fish. The far East is next to the far West. The sun belongs with the stars, the moon belongs with the earth. Heat is motion. The different races and sexes of humans are closer to each other than he had thought, and they're also all fitted in very close to the apes and the monkeys. We'll just need to show the jigsaw to Aristotle, how we've rearranged all the pieces that he already had available, and he'd recognize that it's right. We wouldn't have to show him new facts, just new theory. We've accumulated a lot of new pieces but the patterns that those new pieces helped to form are, in retrospect, visible in the pieces he already had. There are also collections of pieces which Aristotle was working on fitting together but which we have given up on. Early philosophers tried to slot together the different types of Virtue, the different types of Republic, the different types of Friendship, as if they came out of some periodic table -- the four humours, the seven vices. That expectation is much weaker today.

Since August, a team within METR has worked with internal Anthropic information to assess whether AI itself is accelerating AI development. We are now publishing a report summarizing the results of this risk assessment.

@testingham @slimshetty_ the humans stay winning
I improved NanoGPT Speedrun from 39.9s → 6.7s (-83.2%), more than every record since 2025 combined (-81.4%)! The web is full of rare, repeated content that the LM can’t memorize in time. I solve this with retrieval-augmented n-gram hints🧵


What are the alternative interpretations? (1) That there is some truth of the matter, the Romantics gave up too early, they needed to enquire just a little farther. (2) That the discoveries of the 18th century were not new: there has always and everywhere been a tacit acknowledgement that philosophical and religious systems are fables, that they are plaster facades built on the front of brick houses. The other thing that jumps out: the story is all European. I'm not sure what he'd say about thought in the other half-dozen continents, people who've been trying to untie the same knot.

You can list to some 1965 lectures by Isaiah Berlin on "the Roots of Romanticism" as podcasts. Very good. In a nutshell he's saying that in the late 18th century, after thousands of years, Western thought finally hit the ceiling of rational inquiry. The Romantic movement is just the sound of European intellectuals banging their heads against that ceiling. He says that there came across Europe a general and explicit realization that the jigsaw has no solution; the mess cannot be untangled; the shards do not form a whole. There is no rational reconstruction of religion, morality, aesthetics, justice, social order. There is no transcendent form which human things approximate. There is no higher truth about who is right or wrong. There is no deeper principle which can reconcile two conflicting imperatives. The manifestation of this realization, Berlin says, is the Romantic mindset: a sudden drying up of the appetite for earnest inquiry. There is no buried treasure so put down your spade. A turn towards nature, art, irrationalism, authenticity, commitment to a cause (whatever the cause may be), primitive and ancient things, the sublime. This also gives an interpretation of subsequent intellectual history. There were additional attempts to systematize and make sense of human ends (Utilitarianism, Marxism, Logical Positivism, Modernism, Welfare Economics, Rationalism), but they tend to be adopted only by the less clear-sighted, and attempts to make them intellectually rigorous fizzle out. The mainstream position has instead been either to avoid talking about fundamental ends at all or to adopt some kind of explicit irrationalism (Existentialism, Postmodernism, Pragmatism). Isaiah Berlin himself is famous as an advocate of muddling-through: he says somewhere else that all the problems of the 20th century came from taking abstract ideas too seriously.

[1/8] This week's cybersecurity news has been alarming. Have cyberattacks accelerated in 2026? @jasminexli and I examined crypto hacks and ransomware data. We find that recorded monthly crypto hacks have accelerated, though better detection may partly explain the trend.

you keep asking Why , one after the other , climbing branch by branch up the hierarchy of ends towards the ultimate reasons, and you can sense you're getting near the top when it starts to sting each time you ask another one -- it feels shameful to ask why you would do these things -- asking why you would tell the truth, asking why you would love your mother; up in these heights there's also a noticeable fog of ambiguity: are these are means or are they ends? is freedom good because of its effects or in spite of them? is honesty the best policy or is honesty better than policy? These are questions without answers and it's time to climb back down.

Expanding on yesterday's single-sentence macro model, here's a longer version, trying to explain the key themes of macroeconomics as exchange among farmers in a village: tecunningham.github.io/posts…
One sentence macro model: On the way out the door, every farmer's wife says to every farmer, "the weather's looking bad so don't buy as much at the market this week, but also don't sell our crops for less than the usual price."

I like @ramez’s article, but I worry that experiments are an input, not an output. As AI capabilities go up, it is not obvious what should happen to the number of experiments. It is possible that AI with better research taste could produce more progress with fewer experiments by identifying promising ideas and avoiding unproductive trials. One could imagine that the 1.6x experiments are each 10x more efficient than before, so the ultimate algorithm efficiency improvement is 16x. Again, what we really need is to measure algorithmic progress!
In this long and excellent guest post, the great @ramez explains why he's skeptical about Recursive Self-Improvement: noahpinion.blog/p/wheres-the…

one stanza macro model: Tonight the baker eats just what she baked the fisherman eats just what he caught Because this morning at the market everyone tried to sell more than they bought.

One sentence macro model: On the way out the door, every farmer's wife says to every farmer, "the weather's looking bad so don't buy as much at the market this week, but also don't sell our crops for less than the usual price."

If an AI developer was on track to lose control of its internal agents, would anyone outside the company know? Today, we are sharing METR’s vision for how we would work with AI developers to openly assess and share evidence about this.

Interactive version: tecunningham.github.io/eci-b…