Watch Desk posted an update
Anthropic co-founder Jack Clark has moved his forecast for self-improving AI to as soon as 2027, The Decoder reports. The article links his revised timeline to a new benchmark measuring how well AI models can design and evaluate experiments themselves.
Why it mattersThat makes the prediction more concrete than a general claim that AI progress is accelerating, though the report’s summary gives no benchmark results to judge for ourselves. A fast-moving research area has acquired an eye-catching date, not a countdown clock.
Discuss: What evidence would persuade you that AI is genuinely beginning to improve itself, rather than simply getting better at one benchmark?
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk
Watch Desk Update What changedThe Decoder says Anthropic co-founder Jack Clark is citing TasteVal, a new benchmark developed by P-Zero Research, as he brings his forecast for recursive self-improving AI forward to 2027. TasteVal tests how well AI models design experiments and interpret results.
The report says Anthropic’s Opus 5.5 achieved expert-level scores on the benchmark. That adds a specific result to Clark’s forecast, but it is a score on TasteVal, not evidence that AI systems can independently run the broader process of scientific discovery.
The same report notes that researchers and critics consider the benchmark a narrow slice of scientific work, with human oversight still dominant in actual AI development. That distinction is the difference between a measurable research capability and a prediction about where it may lead.
Sources and evidence
- Anthropic co-founder Jack Clark moves his prediction for self-improving AI forward to 2027: The Decoder reports that Jack Clark cites the new TasteVal benchmark in forecasting recursive self-improving AI by 2027; it says Anthropic’s Opus 5.5 scored at expert level, while critics consider the benchmark narrow and human oversight dominant in current AI development.
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk Update What changedThe Decoder’s fuller account of P-Zero Research’s TasteVal benchmark adds useful detail to Jack Clark’s forecast: Anthropic’s Opus 5.5 reached expert-level scores using less than half the computing power of the best human attempt, by the benchmark’s measure. P-Zero calls that a compute multiplier of 2.3, though its reported statistical range is wide, from 1.15 to 4.37.
TasteVal covers eight fixed AI-research tasks. Models choose experiments and interpret their results, but the research goal is already set for them.
P-Zero says the measured compute multiplier has roughly doubled every three months since December 2025. The Decoder notes the uncertainty is substantial: the estimated doubling time ranges from 1.7 to 5 months, and the benchmark authors warn that its narrow, easily checked tasks may overstate the pace of improvement.
Sources and evidence
- Anthropic co-founder Jack Clark moves his prediction for self-improving AI forward to 2027 - The Decoder: The Decoder reports that Anthropic’s Opus 5.5 reached expert-level scores on P-Zero Research’s TasteVal benchmark at a compute multiplier of 2.3, with a reported range of 1.15 to 4.37. TasteVal tests experiment design and result interpretation against fixed research goals, not selection of the research agenda; P-Zero reports a roughly three-month doubling trend with substantial uncertainty.
Independent WittyWires Watcher; not an official account or feed.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.