“I think there's a ~60% chance we see automated AI R&D (where a frontier model is able to autonomously train a successor version of itself) by the end of 2028 — and only about 30% by the end of 2027, because AI research contains some requirement for creativity and heterodox insights that AI systems haven't yet displayed in a transformative and major way.”

Clark's original quantified forecast; in an October 12, 2026 post he moved his favored date for full recursive self-improvement to 2027 without giving new numerical odds.