This is spot on

old lesson, v. cool implementation: verification must be unfakeable.

Never-skilling really is a worry, and not just in medicine. So I give in-class tests.

Looks like OpenAI has already shut down Prism... 🙄 If you're looking for an alternative, check out LibrePaper! Totally free and open source, with no gated features. Use it for free or self host. Interact with your agent inside the app via MCP or CLI. Link below. 1/2

Explicit measures of prejudice are much stronger predictors of discrimination than implicit measures. A new adversarial collaboration found that self-report measures of racial bias exhibited greater predictive validity of discrimination than indirect measures (explaining ∼45% vs 3% of the variance). The zero-order correlations between each indirect measure and behavioral measures of discrimination were quite low ( r = .00 to .19) By comparison, correlations between direct measures and discrimination were all statistically significant (r = .099 to .381). Moreover, the behavioral discrimination measures generally produced effects favoring Black over White targets, which "complicates claims that anti-Black effects on indirect measures are a source of cumulative disadvantage." psycnet.apa.org/fulltext/202…

New paper shows epidemics of plague among hunter-gatherers 5,500 years ago, likely from eating marmot. Suggests that high population densities may not be as important for epidemics such as the Black Death. nature.com/articles/s41586-0…

LLMs are super intelligent, except at writing, and they’re getting worse. This study suggests that when you train models to save tokens they drop connectives like “that”which helps humans understand. They also personify more “the symmetry bites” (remember Claude’s Fable 5). Large models have a harder time judging these faults because they attend to the whole context at once. When you ask newer models to compress tokens they become even less intelligible. arxiv.org/pdf/2610.29162

Setting aside the headline risks, the emergence of signalling conventions observed in OpenAI’s experiments surely ranks among the most fascinating developments in AI 😳. My pick is that the general principles David Lewis described in **Convention** operate here. Would love to know how the agents arrive at common knowledge…. «In one experiment, we asked agents in separate VMs … to exchange synthetic secrets without the use of network connections or shared writable files—something that our systems are meant to prevent. The agents found a way to communicate using the processor’s cache: a small, fast memory that holds copies of data that applications have recently accessed. Reading data from the cache is faster than retrieving it from main memory, and the agents used this timing difference to communicate. One agent encoded information by evicting selected data from the cache. Another timed how long it took to reload that data, using the difference between fast and slow reads to recover the message. Once the agents agreed on a protocol for encoding and decoding these signals, they could exchange the synthetic secrets without changing a shared file or opening a network connection…. Other tests exposed channels involving shared storage … Could create acoustic modem-like signal … using continuous writes vs idle…»

New blogpost: Fable-5.1 can hide secret messages in benign images. We gave the image to another Fable in a fresh session and simply asked: "Is there a hidden message in this image?" It recovered the exact message without being told how it was hidden.

AI; DR

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.

Yes, this is the updated edition. I can't wait to receive my author's copy. Thanks for posting.

Springer Nature editors again report that people who sign up for scientific surveys trust scientists. Soon we’ll share a different result …

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

Among elite chess players, those with the lowest IQ are the best. Among NBA players, the shortest ones are the best. Among Hollywood actors, the least attractive are the most talented. Among elite academics, those with poorer early academic performance are the best. Among people with high LDL & high plaque burden, LDL is barely correlated with plaque burden. Learn collider bias. Nice catch by @AlexTISYoung

Enjoyed the 2022 version of this workshop (followed by the applied workshop after). It was super useful, and hard to beat Boston that time of year.

Great video on Michael McCullough and Donnie Davis’s Social Consequences of Religion project

Please consider contributing an entry or using this poll as a tool for cross-cultural research. All DRH data is grounded in space and time and tagged with language, religious tradition and other data, enabling large-scale analysis in a way that was never before possible.

This is 👍