My weather app said the sky would be clear. I taught an open model to check, then went outside
Description and use
Clear Tonight? uses TabPFN v2 trained on historical cloud forecasts to correct weather app predictions and provide go/no‑go night‑time advice, reducing error by 28–41 % across six cities.
Problem
Weather apps often mispredict cloud cover, leading users to make wrong decisions about going outside at night.
Outcome
Across six cities, Clear Tonight? cut the weather app’s cloud‑forecast error by 28–41%, halved wasted trips in Seattle and Chicago, and produced accurate go/no‑go recommendations within five seconds.
Workflow
- Collect 32 months of forecast and reanalysis data for each city
- Transform each night into a 12‑feature table row
- Label each hour as clear (≤25% cloud cover) or not
- Train TabPFN v2 on the historical rows
- Deploy a live page for six cities and a local CLI for custom locations
- Use Gemma 4 to generate a short plan based on the model’s output
- Log user‑taken sky photos and have Gemma 4 rate cloud cover to build ground truth
Evidence
Clear Tonight? takes the cloud forecast your app shows, looks up 32 months of that same forecast's track record at the same spot, and answers one question with an open model: is the sky really going to be clear tonight, and when should I go?
over a year of nights in 6 cities, it cuts the forecast's error by 28–41%. With only a week of history, TabPFN v2 is the only 12-feature model that beats the app.
The correction cuts the forecast's probability error by 28–41% in every city, and it roughly halves wasted trips
- Record date: 2026-10-11T04:40:56.581Z