The author estimates a 60%+ likelihood that no-human-involved AI R&D — an AI system powerful enough to autonomously build its own successor — occurs by the end of 2028, based on public information from arXiv, bioRxiv, NBER papers, and frontier company deployments.
AI coding ability has advanced dramatically: Claude 2 achieved ~2% on SWE-Bench (a test of real-world software engineering problem-solving) in late 2023, while Claude Mythos Preview now reaches 93.9%. Separately, AI systems' ability to complete tasks independently has grown from ~30 seconds (GPT 3.5 in 2022) to ~6 hours (GPT 5.2 High in 2025) to ~12 hours (Opus 4.6 in 2026), with forecasters expecting ~100 hours by end of 2026.
The author expects a proof-of-concept where a model end-to-end trains its successor within a year or two at the non-frontier model stage, though frontier models may take longer due to cost and human effort involved.
Summaries like this, every morning.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.