Virtual, May 27, 8:00–9:00 PM HKT. Open to anyone regardless of location.
The standard way of answering this question involves benchmarks. OpenAI’s GDPval — built to measure AI on real-world, economically valuable tasks — has already been beaten by today’s frontier models. And yet the real-world economic signals of that automation remain surprisingly muted. Something doesn’t add up.
On May 27, Anson Ho is giving a talk for AI Safety Hong Kong about why that gap exists, and what would tell us something useful instead. Anson is a researcher and writer at Epoch AI. His work covers the economics of transformative AI, the rate of algorithmic progress, US-China AI competition, and the design of AI benchmarks themselves. His Epoch research includes work on whether we’ll run out of training data for LLMs, how fast algorithms in language modelling are improving, compute trends across the history of machine learning, and a recent statistical framework for stitching saturated benchmarks together to track long-run progress. He was also a contributor to the 2026 International AI Safety Report.
His argument for this talk, drawn from a piece on the Epoch AI substack, is that we’ve been searching under the streetlight — tracking what’s easy to measure rather than what would tell us where automation is actually heading. Instead, he’s been trying to get today’s best AI agents to do his own job. Coding tasks, writing tasks, the work of publishing articles. The talk walks through where the agents do well, where they fail in surprisingly human-like ways, and when he expects those bottlenecks to fall.
He’ll also share timelines for coding, writing, and administrative work between 2026 and 2029, and discuss what knowledge-work automation means for Hong Kong — and whether the city’s status as a global financial hub is a strength or a vulnerability in this scenario.
This is the second event in our Speaker Series. No technical background required.
Follow more of Anson’s work here.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.