random_walker Releases ICML 2026 Keynote: What Will Be Left for Us to Work On?

On July 14, @random_walker released the static slides and an annotated, lightly edited transcript of his ICML 2026 Seoul keynote, 'What will be left for us to work on?' The talk takes on a widely watched question: as AI capabilities keep rising and more tasks get automated or markedly accelerated, what work will be left for humans. The materials were quickly reshared and debated across academic and industry circles.

Core framework

Multiple summaries single out 'AI as Normal Technology' as the keynote's central judgment. In @sayashk's retweet, 'normal' does not mean AI is unremarkable; rather, it will reshape society and reorganize labor like past transformative technologies, so it can still be analyzed with familiar methods rather than requiring an abrupt singularity narrative. The framework would only break down, the summaries note, if more radical conditions such as recursive self-improvement emerge.

Reactions

Alongside the slides, the annotated transcript was seen as one of the best entry points to the framework. @math_rachel urged readers to read the full text, cautioning that the talk's discussion of 'automated science' should not be reduced to a slogan, though the post did not elaborate. On the same theme, @No Priors relayed Andrej Karpathy's view that jobs are essentially bundles of tasks, that AI is more likely to markedly speed up some of those tasks, and that it should be treated more as an empowering tool than as a simple prophecy that 'all jobs disappear'.

2026-07-14 ~ 2026-07-15 · 10 related posts