forecastingresearch.substack.com/p/leap-wave-8-ai-timelines
2 corrections found
Among the 264 expert and superforecaster participants who completed both Wave 1 and Wave 8
This participant count is too high. The Wave 8 report says the entire Wave 8 sample included 205 experts and 52 superforecasters, so there were only 257 such respondents total; 264 matched expert/superforecaster respondents is impossible.
Full reasoning
The official Wave 8 report states that this wave included 205 experts and 52 superforecasters. That is 257 expert/superforecaster respondents total in Wave 8.
Because the people who "completed both Wave 1 and Wave 8" must be a subset of Wave 8 respondents, the number of matched expert and superforecaster participants cannot exceed 257. So the article's figure of 264 expert and superforecaster participants is impossible as written.
The report's own wording avoids this number and instead says only: "Among participants who completed both Wave 1 and Wave 8..." That suggests the Substack post likely introduced an erroneous count while summarizing the report.
1 source
- Longitudinal Expert AI Panel — Wave 8: Timelines
The following report summarizes responses from 205 experts, in addition to 52 superforecasters and 601 members of the public collected between Apr 20, 2026 and May 11, 2026. ... Among participants who completed both Wave 1 (Jun-Aug 2025) and Wave 8 (Apr-May 2026) mean expected TRS levels rose...
We defined this as the first year a commercial AI system outperforms the top human performance on the vast majority of non-physical work tasks.
This footnote misstates the survey's AGI definition. The official Wave 8 definition used a 90th-percentile human benchmark, a 90%-of-tasks threshold, and a cost constraint—not 'top human performance' on a vague 'vast majority' of tasks.
Full reasoning
The Wave 8 report gives a more precise definition of the AGI question than this footnote does, and the two are not equivalent.
According to the report, AGI was defined as a commercially available AI system that:
- can outperform the 90th percentile professional human employee in every primarily non-physical occupation,
- on at least 90% of economically useful non-physical tasks, and
- has an inference cost no more than 5× equivalent human labor.
By contrast, the footnote says AGI is when a system outperforms "the top human performance" on "the vast majority" of non-physical work tasks. That is materially different:
- "Top human performance" is stricter than 90th percentile human employee.
- "Vast majority" is vaguer than the explicit 90% of tasks threshold.
- The footnote omits the report's explicit cost constraint.
Because the footnote changes key parts of the stated resolution criteria, it is not an accurate statement of how the survey defined AGI.
1 source
- Longitudinal Expert AI Panel — Wave 8: Timelines
In this survey, we defined “AGI” as a commercially available AI system that meets both of the following criteria: Can outperform the 90th percentile professional human employee in every primarily non-physical occupation (based on 2025 performance), across all sectors, on at least 90% of the economically useful non-physical tasks that they perform; Has an inference cost no more than 5x the cost of equivalent human labor.