The AI Eval Singularity is Near * AI capabilities seem to be doubling every 4-7 months * Humanity's ability to measure capabilities is growing much more slowly * This implies an "eval singularity": a point at which capabilities grow faster than our ability to measure them * It seems like the singularity is ~here in cybersecurity, CBRN, and AI R&D (supporting quotes below) * It's possible that this is temporary, but the people involved seem pretty worried Appendix - quotes on eval saturation Opus 4.6 * "For AI R&D capabilities, we found that Claude Opus 4.6 has saturated most of our automated evaluations, meaning they no longer provide useful evidence for ruling out ASL-4 level autonomy. We report them for completeness, and we will likely discontinue them going forward. Our determination rests primarily on an internal survey of Anthropic staff, in which 0 of 16 participants believed the model could be made into a drop-in replacement for an entry-level researcher with scaffolding and tooling improvements within three months." * "For ASL-4 evaluations [of CBRN], our automated benchmarks are now largely saturated and no longer provide meaningful signal for rule-out (though as stated above, this is not indicative of harm; it simply means we can no longer rule out certain capabilities that may be pre-requisities to a model having ASL-4 capabilities)." * It also saturated ~100% of the cyber evaluations Codex-5.3 * "We are treating this model as High [for cybersecurity], even though we cannot be certain that it actually has these capabilities, because it meets the requirements of each of our canary thresholds and we therefore cannot rule out the possibility that it is in fact Cyber High."

A bit sad to find out that Open Philanthropy’s (now Coefficient Giving) GCR Cause Prioritization team is no more.  I heard it was removed/restructured mid-2025. Seems like most of the people were distributed to other parts of the org. I don't think there were public announcements of this, though it is quite possible I missed something.  I imagine there must have been a bunch of other major changes around Coefficient that aren't yet well understood externally. This caught me a bit off guard.  There don't seem to be many active online artifacts about this team, but I found this hiring post from early 2024, and this previous AMA. 

It seems like a worthwhile project to ask/pressure Anthropic's founders to make their pledges legally binding.  Anthropic's founders have pledged to donate 80% of their wealth. Ozzie Gooen estimates that in a few years this could be worth >$40 billion. As Ozzie writes, adherence to the Giving Pledge (the Gates one) is pretty low: only 36% of deceased original pledgers met the 50% commitment. It's hard to follow through on such commitments, even for (originally) highly morally motivated people.

Potential Animal Welfare intervention: encourage the ASPCA and others to scale up their FAW budget I’ve only recently come to appreciate how large the budgets are for the ASPCA, Humane World (formerly HSUS), and similar large, broad-based animal charities. At a quick (LLM) scan of their public filings, they appear to have a combined annual budget of ~$1Bn, most of which is focused on companion animals. Interestingly, both the ASPCA and Humane World explicitly mention factory farming as one of their areas of concern. Yet, based on available data, it looks like <5% of spending in this category is directed toward factory-farmed animal welfare — despite factory farming accounting for the overwhelming majority of total animal suffering. Given that factory farming is already in scope for these orgs, and that is responsible for the vast majority of animal suffering, it would seem quite reasonable for these orgs to increase their spending on FAW several-fold. I doubt their donors would object!

Lots of “entry-level” jobs require applicants to have significant prior experience. This seems like a catch-22: if entry-level positions require experience, how are you supposed to get the experience in the first place? Needless to say, this can be frustrating. But we don’t think this is (quite) as paradoxical as it sounds, for two main reasons.  1: Listed requirements usually aren't as rigid as they seem. Employers usually expect that candidates won’t meet all of the “essential” criteria. These are often more of a wish list than an exhaustive list of strict requirements. Because of this, you shouldn’t necessarily count yourself out because you fall a little short on the listed experience requirements. Orgs within EA are much better at communicating this explicitly, but it should be taken as a rule of thumb outside of EA as well. You should still think strategically about which roles you apply for, but this is something to factor in. 2: You can develop experience outside of conventional jobs. For a hiring manager, length of experience is a useful heuristic. It tells them you’ve probably picked up the skills needed for the role. But if you can show that you have these skills through other means, the exact amount of experience you have becomes far less important. A few of the best ways to do this: * Internships and fellowships. These are designed for people entering new fields and signal to employers that someone has already vetted you. They’re often competitive, but usually don’t require previous experience. * Volunteering. Organizations usually have lower bars for volunteers than paid positions, making this a more accessible option (usually). Look for advertised volunteering opportunities at orgs you’re interested in, or reach out to them directly. * Independent projects. Use your spare time to make something tangible you can show potential employers, like an app, portfolio, research paper, blog, or running an event. Obviously the most useful projects will v