Import AI 477: In which I tell the IMF the Singularity has begun
Import AI Jack Clark
At an IMF meeting in Bangkok, a tech veteran said AI has hit the early singularity. He says it’s already changing science, security, and growth faster than policymakers expect.
Based on reporting by Import AI, Jack Clark — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
At the IMF meeting in Bangkok on Saturday, October 17, the speaker made a blunt case: AI has already crossed into the early singularity. He used the Per Jacobsson lecture to tell economists, bankers, and policymakers to brace for rapid change, not someday but now. The point of the speech was not just that AI is impressive. It was that it is starting to shape science itself, sometimes on its own.
He framed the moment through his own career arc. He started as a journalist covering data centers, moved to OpenAI in 2016, and later co-founded Anthropic. That path, he argued, gave him a front-row seat to a shift that now touches national security, the global economy, and the machinery of research. He said AI is already helping to develop itself, and that this should be treated as a sign of much larger change ahead.
The strongest evidence, in his view, is in science. He pointed to AI systems that have proposed solutions to hard math problems, helped design viruses from scratch, identified drug candidates, and even helped create drugs now in trials. In robotics, he said AI is increasingly able to operate lab equipment remotely, which could speed up experiments in biology, materials science, batteries, and fuel cells. He tied this to a broader claim: the same frontier model can now work across multiple disciplines.
He leaned heavily on benchmark scores to make the point. Humanity’s Last Exam launched in January 2025 with top systems at 8%; he said the highest score this week is 67.7% from Claude Opus 5.5. FrontierMath, which launched in November 2024, went from 2% at the start to 94% now. And then there was TheoPhysEval, a paper from last week on unpublished theoretical physics problems, where he said GPT-6-Astra ended up at 98% after the model itself flagged errors in expert answers.
His larger argument is that this is what frontier science looks like when AI starts to contribute creatively: the pace of progress changes, the pool of “researchers” can grow dramatically, and countries that prepare may benefit most. He thinks the next few years will bring not just more capable systems, but a very different rate of discovery.
My take — AI-written commentary, not fact-checked reporting
This is the sort of speech that makes policymakers reach for the aspirin. The awkward truth is that the AI industry has moved from hype to leverage, and now wants everyone else to plan around it. That’s fine, but the people selling speed rarely volunteer to pay for the clean-up.
Read more about this at: Import AI