CNN-based ENSO model called a very strong El Niño months before physics models
4billionyearson · reddit · 2026-08-04
A Reddit post highlights a CNN-based ENSO forecast that called the current El Niño “very strong” months before physics-based models caught up.
Key claims in the post:
- Seoul National University’s convolutional network forecast a very strong El Niño back in April 2026.
- NOAA’s dynamical models only converged toward that level months later.
- The CNN learns ocean-atmosphere behavior directly from decades of observations rather than simulating physics.
- It reportedly retains forecast skill 18–24 months out, past the usual spring predictability barrier.
- The model now also projects a flip to La Niña in 2027, a horizon physics models cannot reach.
The post links to a longer write-up and a live tracker comparing AI and physics forecasts week by week.
More from Research
- Paper maps 41 agent failure modes to model, harness, tools and environment — omarsar0 · 2026-08-04
- Claude thread proposes an adversarial review skill for agentic science — ChrisGPotts · 2026-08-04
- Singular learning theory talk links neural-network generalization to alignment — andrewgwils · 2026-08-04
- AI paper argues best-of-K boosts generative expressivity, not just sampling quality — anshulkundaje · 2026-08-04
- ASCII art may be a better taste benchmark for frontier models than you think — weswinder · 2026-08-04
- A curated reading list for DeltaNet, FlashKDA, vLLM serving and MoE — austinvhuang · 2026-08-04