iiinova — where your digital self keeps evolving

Seth Wynes

@sethwynes.bsky.social

774 followers 139 following 22 posts

Assistant Professor at the University of Waterloo. Climate, policy, housing, AI risks.

Seth Wynes @sethwynes.bsky.social · 22/01/2026

New preprint: Climate Research Agendas Should Account for Anticipated AI Risks. I argue that AI timelines are shorter than climate timelines and since climate research is unique in focusing on such long horizons, this requires some adjustments from climate folks.

Global surface temperature change relative to pre-industrial times (A), and task duration that state-of-the-art AI systems can complete with 50% reliability (B). Note the logarithmic y-axis in Panel B. Task duration estimates are from METR (Kwa et al., 2025). Forecasted artificial general intelligence (AGI) range includes lower 25%, median, and upper 75% estimates from forecasting platform Metaculus as of December 2025. Climate data from IPCC AR6 Synthesis Report SPM.4 (a) and Figure 3.3(a)(IPCC, 2024).

100

Seth Wynes @sethwynes.bsky.social · 12/09/2024

We also asked these climate experts to estimate the beliefs of their peers. There were strong correlations on all questions, suggesting that participants with very high or low estimates perceive their own beliefs as closer to the community average than they actually are.

100

Seth Wynes @sethwynes.bsky.social · 12/09/2024

We expected that researchers who worked on solutions (WG3) might be more optimistic than those working on adaptation (WG2), but found limited support for this hypothesis (differences were only significant between WG2 and WG3 for estimates of the likelihood of reaching 3°C)

100

Seth Wynes @sethwynes.bsky.social · 12/09/2024

Instead, the median estimate of maximum warming by 2100 was 2.7°C, which is consistent with modeling of the climate response to current national climate policies.

100