Day 71 of 100 Days of AI
Today, I continued testing my crew of agents that will find and summarise product reviews. I probably won’t have anything publicly ready until the weekend.
Meanwhile I came across this interesting article about how deep learning is being used to predict the weather.
Traditional systems in use today rely on complex physics modelling. You need massive compute for these models to run. However, with deep learning, you can train a neural net on historic weather data and run it on a single desktop computer with a powerful GPU.
How are these new AI weather models trained? Here’s an excerpt from the article.
One thing I got to appreciate from the AI courses I’ve completed in the last 2+ months is that machine learning has some major limits. And even though these weather AI models might be useful in some cases (and cheaper to run!), unlike the traditional physics modelling, they rely on historical data that isn’t always indicative of the future. Here’s a noteworthy comment from Hacker News that flags this issue:
