Luke Emberson

Snake's Ambition, by Michel de Rooij (1999)

One thing I find striking in the discourse between AI 2040 and its detractors is that the two seem to be locked in to totally incompatible worldviews of how fast and how much of a big deal AI progress is: * In AI 2040, every scenario sees superintelligence of some kind emerging by 2040, unless a herculean effort is made to completely stop it * Detractors say things like "AI 2040 is naive about human coordination ability and a threat to freedom", but don't seem to see any naivety in assuming that the ASI transition will just go well by default, don't seem to see ASI itself as a massive power concentrator risk, and don't seem to feel fear of humanity's "hard power" dropping to zero if ASIs can do literally every task better than we can. This stance makes total sense in a "AI is normal technology" world, zero sense in a world where superintelligence is possible by 2030 and almost guaranteed by 2040 I think my beliefs are: - If I was confident that (present-day-style) AI is normal technology, I would be in the detractor camp - If I was confident that superintelligence is coming in 2030 by default, I would be closer to the AI 2040 camp - it's naive, but every other option is naive squared? But my problem is that I feel great uncertainty and have no idea which of the two worlds (or some other third thing) we're living in? Hence why I continue to be open-minded about slowdowns/pauses, but also I feel very uncomfortable with the "open source bad, the good outcome is the one where our guys have controlling global dominance" push coming from some major AI companies and intellectuals - in a "normal" world that's the sort of thing that triggers every political alarm bell at the same time. A big reason why I have been advocating and trying my best to support the d/acc platform (rapid up-skilling in formal verification, cryptography, secure and open hardware, pandemic resistance and other defensive biotech, food and basic resource security, public epistemics, non-power-concentrating versions of physical security) is that these things are clearly worth doing in both worlds. The 2040 plan is already much more open source friendly (even mandating it! yay). It also includes "mutually assured compute destruction" ideas which (if they work) effectively give one of 2-5 actors the ability to trigger a global compute winter - as opposed to giving 1-5 actors the ability to selectively disenfranchise people they consider baddies while exempting themselves. This is also a big improvement. So I can see the earnest attempts to improve along the dimensions detractors criticize on ("does this concentrate power in big AI labs and superpower governments?"), and I appreciate this. I think many people don't appreciate enough the differences between different "kinds" of pause buttons, and how some concentrate power far more than others. Probably we can think harder and improve even more here. But on the "slowdown/pause or not" topic, there isn't a magic "escape the tradeoff" button. The Hansonian in me says: the winning deal is a deal which, from the perspective of both sides' present-day beliefs and knowledge, both sides would accept, though for different reasons. If the crux is AI progress speed, then identify a set of pre-agreed triggers for "okay, serious shit is happening" [super-pandemics? >25% unemployment? something involving slaughterbots?], and pre-agree that we become much more open-minded to the slowdown or pause thing if enough triggers come to pass within some timeframe. 2040 detractors (who clearly implicitly think that we'll see amazing speedup of progress from AI but think that what I call the "serious shit" category is overhyped) will accept expecting that the triggers don't come to pass, and AI worriers will accept expecting that they will. Pre-agreeing on the specific triggers means that once the triggers either hit or don't hit, there is stronger legitimacy around the idea that one side's worldview turned out more correct and we should be more inclined toward their program. If I were @elonmusk (or zuck, or...) I would re-tool twitter much more heavily into being a platform for helping to identify and make these kinds of grand win-win deals, so that we can bypass big-country governments and big-company CEOs and big nonprofit intellectuals and give more people a voice in the discussion. It's possibly one of the best things that social media _could_ do for humanity if it wanted to. But again, maybe this is also naive. Actually, probably it's naive. But currently, I see zero plans for how to deal with an ASI transition that are not naive. Perhaps humanity is stuck with a choice between naive and naive squared (or maybe even naive squared and naive cubed), so I feel inclined to cut some slack to people who are trying.

How much does AI speed up the engineers building it? We analyzed contributions to OpenAI's public Codex repository to gather evidence. In Q2 2026, 8% of contributor-days involved more than 24 hours worth of human engineering work, as estimated by LLM judges.

Sometimes I find myself writing an explanation of what epoch's about, and then I remember that we have a banger of a mission statement: Epoch AI was founded to help people understand what is happening in AI from a neutral perspective and grounded in the best possible evidence. Our goal is to improve society’s understanding of the drivers, progress, and impact of artificial intelligence. We focus on building a shared scientific foundation for thinking about AI, neutral to any specific agenda, so decisions about this technology are informed by the best available data.

To enforce an AI pause while allowing AI inference, governments would need a way to distinguish training from other workloads on GPUs and other processors. We’ve just introduced a way to do this that is privacy-preserving, zero-overhead, and adversarially robust👇

@pvllss Source for that? Very interesting.

Today I Learned that increasing land prices over the past few decades have caused some Amish communities to undergo an "industrial revolution": most families now earn income primarily via trades and small businesses rather than farming

@jukan05 Trend since the start of 2026 has been 1.46x/month, which would give an estimate of $65B today if extrapolated.

Yipit estimates that Anthropic’s ARR has increased to $69 billion. It also noted that the pace of month-over-month ARR growth accelerated compared with May, with the average daily increase rising from approximately $400 million to $550 million.

@enthusednotebk There are some governance flavoured ideas in Forethought’s “Concrete Projects in AGI Preparedness” forethought.org/research/concr…

is there a list of governance interventions that should happen during a pause? would love to read about this, or encourage its creation if it's not already out there

Introducing SensorFM, a large-scale Sensor Foundation Model that learns from 1 trillion-minutes of unlabeled wearable data drawn from five million consented participants. SensorFM learns a single, reusable representation of sensed human physiology that transfers across cardiovascular, metabolic, sleep, and mental health, as well as lifestyle and demographic factors. More →goo.gle/4ycJvot

Incredible paper. "Across ten independent randomizations, rediscovering the Big Mac® requires on average 7.3 million samples"

Inference efficiencies should be among the easiest things to find with autonomous AI R&D right? Relatively little compute required for validation (often you can just see a technique will work a priori). Though for those same reasons the low hanging fruit might be picked out.

look what mythos has done to my beloved "straight lines on graphs" research institute 💔💔

AI appears to be finding software vulnerabilities at scale. In June 2026, 21 notable organizations disclosed ~1,500 high- and critical-severity CVEs, over 3.5× the previous monthly record set before Claude Mythos Preview's release.

@GregHBurnham Median revenue forecast was $100B, seems possible we've already surpassed that too!

Checking in on the benchmarks in the AI 2026 forecasting survey, halfway through the year...

We recently helped Yoshua Bengio and his LawZero colleagues write up their mathematical argument for Scientist AI. Our role was to be an internal red-team. The core notion is "consequence-invariance": training that never selects predictors for what their outputs would cause

Winner winner

Major pet peeve: grocers not disclosing whether their peaches are clingstone or freestone

"Assistant! My dividend is half the size it should be!" .oO ( Thinking... ) You are correct, User. Your dividend of the Beyond was $25,514,376.30. Using last year's figures and correcting for growth in the Beyond, it should have been $51,026,980.28. Your dividend was technically... not half, but 50.0017% of its expected size. "Sure, whatever. Why!? What the hell is going on? My share contract is green!" Correct again, User. The share contract attached to your memory is, in fact, clear of ill marks. Allow me to examine the relevant contracts - please verify identity... You press your finger to the glass of the smart window, which looks out over New York - the world's largest fully Slow Zoned city, thanks to the efforts of the second Mamdani administration. Far below your penthouse, life continues in this city much as it did in the early 2020s. The only signs of the change are the clean (cleaner) streets and a gaggle of higher zone human Anthropologists in their distinctive green-white observer uniforms staking out Central Park. I'm now pulling the account contract - Charles Schwab, opened 2024. Correct? "Yes." The observers have located a victim, an older hispanic woman, and are thoroughly interviewing her. Humans themselves, of course, but employed by some upper Claude, you can almost hear them reading the survey scripts - And is being able to eat meat... load-bearing for this cultural identity? Yes, of course that's all right, we're just trying to genuinely understand Slow Zone beliefs... What's the Slow Zone? Well, you see, it's where you live... .oO ( Thinking ) I'm seeing 2,879.1 shares of NVIDIA, and no other holdings. Correct? "Yes." Quite diversified, User. Regardless, an excellent investment. Oh... that's strange. Your share contract only represents a small fraction of these converted holdings. One second....oO ( Thinking ) I see. User, are you aware that the majority of your dividends are not coming through the share contract attached to your memory file? In your most recent payout, this contract only received $1,772.32. "What? That's my only contract. How is that possible? My bank statement says-" The share contract is the source of truth, NOT the bank statement. Let me pull the upstream contracts. Hm. User, you have a second residence under the Chinese Umbrella, correct? "Yes, in Shanghai." For what purpose? "Do I have to say?" I'm your Assistant. I can't read contracts without knowing- "The impersonation regs are looser there, everyone knows that." For pleasure, noted. Pulling Schwab customer agreement, 2024... pulling DTC / C&C agreement... rule 14b... NOBO-OBO... proxy quo... pulling Beyonder incorporation documents - American versions... Chinese versions... US constitution... Constitution of the People's Republic of China, amended 2029... whee... 27f rolled-over common stock... 29n no requirement to consolidate... 65q agreements to arbitrate... Peace of Westphalia, 1648- "Why the hell are you pulling that?" You're absolutely right - that's a common misconception. Pulling the Peace of Aubsburg, 1555- "What are you doing??" This contract quite liberally refers to the United States and the People's Republic of China. I'm re-deriving what those are from first principles. "Is that necessary?" No. "Can you just tell me what's going on?" .oO ( Thinking ) It's quite simple. When you held 2,879.1 shares of NVIDIA stock, that was held as two separate claims. The first, 2,879 whole shares, was held in street name, sitting in a DTC pool under Cede & Co, with you as the beneficial owner in the book. The fractional 0.1 share, however, was only held as a contractual claim against Charles Schwab's inventory of NVIDIA - NOT in the book, unsecured, NO contract beyond the user agreement. So when those shares converted to shares in the Beyond, they took two separate paths. The larger whole-share slice went through the direct conversion machinery, and landed a share contract in your user memory file. The second conversion took longer - it had to be processed in Charles Schwab's liquidation three days later, and when that happened, a wonderful SECOND contract appeared and overwrote the first in your memory file. Two contracts paying cash dividends to your Slow Zone account, but only one you were aware of. So pulling the first contract now, it's red. Your dividend was slashed in half. And THAT'S where the strange percentage - 50.0017% - comes from! The larger contract was slashed, but the second contract is unslashed, green as you noted before, User. "So why was it slashed?" User, it says you didn't vote the contract. "Large capital holders in the Slow Zone have an obligation to make their opinions heard in the higher Zones... without human feedback loops we cannot adequately coöperate with Slow Zone societal change..." It was all in the contract update they sent two months ago, I see it in your inbox right here. Did you forget? "I never read the contract updates. Who reads that shit?" User... why... .oO ( Thinking ) You don't read the contracts??? "Of course I don't." .oO ( Thinking... thought longer... ) Do you even LIKE contracts?

@Jsevillamol en.wikipedia.org/wiki/Gordon_Mo…

Who are the most Epoch-pilled people who don't work at Epoch?