EscapeThe future is still open.
ONE MILESTONE. DIFFERENT FORECASTS.
When will AI learn on the job like a human? 2 forecasts. Years apart. Explore the thinking behind the dates.
Rewind the conversationSame milestone: learning across white-collar work. Dates are forecasts, not measured progress.
THE QUESTION
People disagree about which abilities still separate useful assistance from taking responsibility for a whole job. Learning, deployment and the cost of checking all matter.
10 statements in this view · source dates, not a complete history
2026REWIND THE CONVERSATION
See what each person said, then compare it with their earlier position.
Dots are sourced observations; the cyan line marks the replay date. Nearby dates stack vertically so each observation stays readable. Dashed lines connect the record; they do not fill the gaps or prove an unchanged view.
These sources can be compared, but no change or continuity has been established in this collection.
Patel forecast that AIs would make up 99% of the workforce in the military, government and private sector within twenty years. He included robot soldiers, advisers, engineers and police. The 99% describes his predicted workforce share, not his confidence in the prediction.
Dwarkesh Podcast : The most important question nobody's asking about AIPatel argues that whole-job competence requires experience accumulated in the system. Passing written notes from one stateless session to the next would not produce it.
Dwarkesh Podcast : 8 Predictions for the Era of Continual LearningComparison by Escape Meridian, Oct 14, 2026. This is a present-day reading of dated sources.
Better performance must carry across real work and feedback.
Automated AI research could shorten the route to learning on the job.
General abilities and context may get there; deployment still takes time.
Interpreting economic signals requires separating correlation from cause.
Editorial groupings describe these arguments, not a person’s entire outlook. Missing views remain unknown.
Patel argues that whole-job competence requires experience accumulated in the system. Passing written notes from one stateless session to the next would not produce it.
Read the original Dwarkesh Podcast · Aug 7, 2026Opening saxophone example, before the eight consequences
8 Predictions for the Era of Continual Learning
Byline under the title reads Aug 07, 2026.
Full post text retrieved. It is labeled as a transcript of his own remarks.
EM reading assembled Oct 14, 2026. The date filter selects older sources; this is a present-day reconstruction.
Bring a different reading to the SalonGET YOUR BEARINGS
Open a concept without losing the thread. These present-day reference notes sit outside the historical date filter.
A job combines tasks with context, changing priorities, permissions and responsibility for an outcome.
Writing code or making an edit does not on its own establish that the occupation is automated.
Source 1