We just launched the next iteration of Precip.
The most accurate minute-by-minute rain forecast in the US:
• 52% better than Google
• 85% better than Apple Weather
The best weather app since DarkSky.
1/ Today we’re unveiling Bombadil, @electricplantco's foundation model that reads the electrical signals of living plants and turns them into a continuous view of their physiology.
Trained on 7M+ hours of plant electrical signal across 125+ species. That makes Bombadil the most species-diverse model of directly recorded electrical biosignals in existence. The animal equivalent would be one model that reads the brainwaves of blue whales, horseshoe crabs and fruit flies.
@RSarava@danrothenberg Yeah, and try comparing either to URMA or gauges. They are not representative of actual rainfall. Far from it. Thanks for sharing the comparison.
@danrothenberg It is especially harmful to measure the foundation models against ERA-5 because that what most of them trained on. They inherited the low skill from the low quality training data. This verification would reward them for making a bad prediction if it matches the bad target.
@danrothenberg I want to love your OWB dashboard, but I just can't understand how to use it when the target is the model's own analysis. Even worse, ERA-5 for precipitation is such an inaccurate target that it would be better to not even attempt verification at all. What am I missing?
@shreya_68@shoyer Great question. We're running inference on live data (about 2-minutes old) on @nvidia GH200 hardware. One 4-member ensemble run takes about 5 minutes for North America with 5 minute steps out to t+120 minutes. Available commercially in our API and for free with a 24 hour delay.
We promised receipts on our new rainfall nowcast. Here they are.
The full validation report is now available, breaking down our accuracy across three separate benchmarks using numerous scoring methods.
Links to the interactive version and PDF whitepaper in the comments.
@bapadadada Moderate to heavy rain rates is where models struggle the most so we're quite impressed that ours performs well here. BTW - Gemini considers 8mm/hour as moderate not heavy and I agree.
@NickHorob For sure, but we we're monetizing just fine until Monsanto decided to push out Climate for free with incredibly aggressive anti competitive practices. That changed everything.
Precip raised a small seed round from some incredible angel investors, and is growing on cash flow.