Can You Trust Machine Learning Chat Applications for Weather Information?

From the Cliff Mass Weather Blog

Cliff Mass

Today, machine learning (ML) applications such as chatGPT are all the rage, with folks worried about the displacement of humans, bogus homework, and more.   

So I was not a little curious to evaluate OpenAI’s chatGPT abilities in the meteorological realm.

My conclusion:  human meteorologists currently don’t have much to worry about.

ChatGPT can produce reasonable-sounding prose that is often totally wrong.

For example, I asked it about the origin of the all-important Puget Sound convergence zone (see below).

According to chatGPT convergence zones occur when ocean air collides with drier air from the eastern part of Washington State.  Wrong.  The Puget Sound Convergence occurs when marine air moves around the Olympics and then converges (comes together) east of the Olympics, forcing air upwards (and thus producing clouds and precipitation).   Zero for chatGPT

Then I thought I would give it an easy one…why Bellingham is often cold.  The answer is that cold air from the interior can jet out of the Fraser River Valley, a low-level passage through the Cascades.  ChatGPT came up with a crazy answer having to do with rain shadowing by local mountain ranges.  Totally bogus.

Then I asked about why there is often cold air in the Columbia Gorge.  Another easy one.

Truth: the low-level Columbia Gorge acts as a near-sea level conduit of cold air from eastern Washington.   But chatGPT had other ideas, from elevation (it got it 180° wrong), rain shadowing, and cold air drainage from the slopes of the Gorge.  Very bad explanation.

In sheer frustration, I decided to give chatGPT an even easier question: to produce my bio.

Much of it was wrong.  It claims I was born in 1951 (wrong) in Brooklyn (wrong) and that I grew up in Great Neck, NY. (wrong).  It was I got my Ph.D. from MIT (WRONG…I graduated from the UW).   And many of the other “facts” were in error.

In summary, chatGPT’s performance was generally quite poor and one has to be VERY careful before believing its often convincing prose.

Finally, I asked chatGPT to write a weather blog for me and it was happy to oblige (see below).  I will let you, my readers, decide whether it is good enough to take over.

A lot of it was wrong, by the way.

via Watts Up With That?

https://ift.tt/nP57ijQ

February 16, 2023 at 05:00PM

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s