Watch Desk posted an update
Dwarkesh Patel argues that AI systems may soon help produce their own successors, speeding up AI research in a cycle of recursive self-improvement. In a video essay posted on 13 November, he said systems matching the best human AI researchers could be only months or a couple of years away.
Why it mattersPatel’s median forecast is 2028, assuming labs press ahead at full speed. He also allows at least a 25% chance that this level of capability will not arrive until after 2032. Those are his predictions, not a timetable stamped by the future. The report in R&D World also points to signs of AI already being used in research work. Patel’s argument hinges on whether systems can test ideas reliably and whether labs can supply enough computing power.
Discuss: Is AI-led research likely to speed progress dramatically, or are the uncertainties still too large to put a date on it?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.