Levin describes research spanning AGI forecasting, safety and scientific applications. He argues that useful benchmarks must track the capability frontier and calls for better measurements of long-horizon tasks, sample efficiency and world models.
He connects Kurzweil's forecasts to improving compute economics and attributes recent mathematical progress to compositional reasoning and readily checked proofs. He distinguishes success in mathematics from competence across the full range of human cognitive tasks.
The interview then considers alignment, model behavior and public attention to risks. Levin presents 2029 as his probabilistic forecast, not a demonstrated arrival date, and argues that additional algorithmic or architectural advances remain necessary.
Watch on YouTube
