Uh. Wow! ChatGPT

[Michael Wilkinson]

I just told ChatGPT to write a song about a stormchaser.

Verse 1: As the clouds roll in and the winds start to rise,
The stormchaser's heart begins to ignite.
He packs up his gear and hits the road,
Chasing the storm, his ultimate goal.

Chorus: Oh, stormchaser, fearless and bold
Racing across the landscape, uncontrolled.
Seeking the thrill of the raging storm,
He's always chasing, never warm.

Verse 2: The lightning flashes, the thunder roars,
The stormchaser's senses are heightened and raw.
He braves the gusts and the driving rain, Determined to capture it all, the pain
and the strain.

Chorus: Oh, stormchaser, fearless and bold,
Racing across the landscape, uncontrolled.
Seeking the thrill of the raging storm,
He's always chasing, never warm.

Bridge: He's a thrill seeker, an adrenaline junkie,
Chasing the storm, never feeling unlucky.
With a heart full of courage and a mind full of fire,
He's always chasing, never tired.

Chorus: Oh, stormchaser, fearless and bold,
Racing across the landscape, uncontrolled.
Seeking the thrill of the raging storm,
He's always chasing, never warm.

 

[James K]

Nice.

There are apparently AI systems where you can give it text lyrics, & a style, then it'll generate a 'song' from that...
This needs to be tried in one of those!!
(I think they're pretty limited in length though (might have to do it in parts)? and sometimes turn the text you give into gibberish)

 

[Michael Wilkinson]

I think there’’s a little bit of gibberish in this one too. I can’t come up with a meaning for “never warm” at the end of the chorus. I think it needs some tweaking. We could probably somehow replace that line and put the word “warned” in there at the end of the chorus. Someone could probably come up with a creative way to make “warned” mean something describing the chaser while also making it a double-meaning with a play on a tornado warned storm.

I’m still impressed with what it did on the first try.

 

[Michael Wilkinson]

Or, quite literally use “tornado warned”.


Chorus: Oh, stormchaser, fearless and bold,
Racing across the landscape, uncontrolled.
Seeking the thrill of the raging storm,
Finding the one that will go tornado warned.

 

[JamesCaruso]

Good one, I would suggest one little tweak, change “will go” tornado warned to “becomes” tornado warned

The bigger OT issue here: Anybody else fearful of this technology? Obviously as a chaser, and a corporate finance executive in my day job, I am no Luddite. But there is something troubling about eventually being unable to know whether any sort of content - music, writing, research papers, student papers in a university setting, even audio or video passed off as news - are authentic human creations, or AI-generated. People or organizations can pass off AI content as their own. One might say that the gibberish is a giveaway, but we all know that will eventually diminish as the technology improves.

 

[rdale]

Academic papers are not a worry at this point because it can't cite, plus it still spits out enough gibberish when you ask info on an academic topic that you know it's not real.

If the art or music or fictional books it creates are good - then go for it!

 

[James K]

I do mess around with the AI art stuff, and all I can say is its impressive. Sure you can make crazy stuff, but I've also seen plenty of images that look basically like a real photo. The one spot where the AI-art systems fail completely is if they put any text on the image - its always pure gibberish.

I haven't tried ChatGPT or any other 'text' ones, but the above is pretty impressive (not just because its fairly clear, but it also has the general idea of a stormchaser..)

 

[Cameron E.]

Verse 1: As the clouds roll in and the winds start to rise,
The stormchaser's heart begins to ignite.
He packs up his gear and hits the road,
Chasing the storm, his ultimate goal.

Chorus: Oh, stormchaser, fearless and bold
Racing across the landscape, uncontrolled.
Seeking the thrill of the raging storm,
He's always chasing, never warm.

Verse 2: The lightning flashes, the thunder roars,
The stormchaser's senses are heightened and raw.
He braves the gusts and the driving rain, Determined to capture it all, the pain
and the strain.

Chorus: Oh, stormchaser, fearless and bold,
Racing across the landscape, uncontrolled.
Seeking the thrill of the raging storm,
He's always chasing, never warm.

Bridge: He's a thrill seeker, an adrenaline junkie,
Chasing the storm, never feeling unlucky.
With a heart full of courage and a mind full of fire,
He's always chasing, never tired.

Chorus: Oh, stormchaser, fearless and bold,
Racing across the landscape, uncontrolled.
Seeking the thrill of the raging storm,
He's always chasing, never warm.

This is a Hank Williams Jr. kinda song for sure haha.

 

[JamesCaruso]

Academic papers are not a worry at this point because it can't cite, plus it still spits out enough gibberish when you ask info on an academic topic that you know it's not real.

If the art or music or fictional books it creates are good - then go for it!

I‘m worried more about where this is clearly headed, not just where it is today… If AI can’t do citations now, I have to imagine that’s not far off. Seems relatively simple, given that AI could easily search the academic literature in the public domain and indicate the source. If it can’t do that today, I’m not sure why, but either way it can’t be far off. As far as gibberish, like I said in my post, yeah that’s a giveaway now but we all know it is going to become less and less a factor as the technology improves.

I also think art, music and books are a bigger issue than what you seem to imply, i.e. “go for it,” by which I assume you mean just enjoy the output for what it is… For me there would be a distinct difference between reading something profound, written by a human author, feeling that empathy and connection even though you never met the author, versus reading something that resonates with you but leaves you with the nagging thought that it may have been written by AI. Imagine, for example, reading a human chase account, empathizing with a fellow chaser’s frustrations, only to find it’s an entirely fictional account created by AI. Take it a step further, and imagine a digital image of a tornado that never happened, created by AI and made to appear as if it’s a real event that nobody else saw. If it’s artificial, it’s just not the same. It’s the reason a reprint is just not as valuable as the original art created by the painter’s hand. Or why an artificial plant is just not the same as a real one, no matter how “real” it looks. Or why I wouldn’t want to replace a real dog with a robotic version.

There is still much to be lost to AI in the human experience, no matter how high the quality, efficiency, or even beauty of the output…

 

[Skip Talbot]

Anybody else fearful of this technology?

Nope. It's just another bit of tech and a new tool like all the rest. If the AI writing term papers and taking college exams becomes a problem, then it's time to reassess the value of those papers and exams. Just like rote memorization and recitation verbatim aren't great ways of demonstrating if a person has a deep understanding of conceptual meanings, if an AI can just fetch bits of text and string them together to make passable college assignments, then maybe those assignments aren't really that useful anyway.

If automated content creation creates a crisis in authentication and validity of information, then guess what? It's time to reassess how that's done too. Frankly, we're looong overdue for an overhaul in information verification and critical thinking. The idea that any collection of words or pixels is "real" or "true" because a person or camera made it is laughably basic. That "fake news" is somehow new or more prevalent because we have new transmission channels like the internet is ridiculous. We've been spreading unvetted misinformation for thousands of years. I'm looking at you Holy Books. Oh, I don't mean *your* Holy Book. I mean all those other ones that are obviously false. Information will need a line of evidence behind it that anyone can quickly follow, a provenance, and integration into other verifiable information. "Real" or "person created" things might need something "wet" attached to them before they can be deemed authentic, if that's even needed. Otherwise folks will learn (one way or another) that in the absence of those things, all they have is merely a pile of words and pixels and nothing more.

Music, art or literature only need only sound or look "good" or evoke emotion to be valuable. Humans have also been stringing together bits of incoherent gibberish for thousands of years only for it to become a "smash hit". That an AI can also do that isn't that surprising. If the value in music and art is instead its "realness", then folks need to go to live performances or sessions where it's created in person.

If a ChatGPT successor actually learns how to cite scientific papers, use interconnected logic and higher math (it currently can't) on actual data, and produce a research paper that can not only pass peer review but is indistinguishable from human generated papers, then as far as I'm concerned, it's doing real science and should be welcomed. We generate so much data now, there's no way we could ever possibly sift through it all. Machine Learning and AI will become necessary tools to science.

As always, I think the fear with this emerging tech is that people don't yet understand what major limitations it has, and what role it will play as a useful tool. ChatGPT isn't AGI like Skynet. It's a glorified parrot. This is just like what people thought of digital computers and robotics, demonstrated in the popular scifi tropes of the 1950's, 60's, 70's... :

"I am now telling ChatGPT exactly what it can do with a lifetime supply of chocolate."

 

[Jeff Wright]

Have it combine the Tuscaloosa tornado of 2011 as The Dunwich Horror…rushing airy presences…a squirming mass of ropes…

 

[James K]

If AI can’t do citations now, I have to imagine that’s not far off.

There is one out there that can (don't remember what its called) but at this point its just citing stuff like Wikipedia.

If the AI writing term papers and taking college exams becomes a problem, then it's time to reassess the value of those papers and exams.

Fact is that apparently is starting to become an issue, its something college instructors have to watch out for on term papers.

 

[JamesCaruso]

@Skip Talbot Not sure I agree with all of your points, but you definitely offered some alternative (and more optimistic) ways to think about AI. I appreciate your insights and longer-form, thoughtful post.

 

[Jonathan Scrogham]

I think it's no longer a question of if AI can do this, but should AI do this. It's more of an ethical issue at this point, and I'm not sure how I feel about it.

 

[Jason N]

Some co-workers of mine and I were talking about this at work today and the topic of, "does this technology hinder or promote innovation" came up. I think if you are asking it to reach out into the abyss of information to produce content, research, or information, do you have to "train" or "aim" it to stay away from certain sources? and does that end up being a self-licking ice cream cone in the sense of, "it's producing what I asked for faster, formatting it, but is the information actually good and am I going to end up doing more work or less work to make it "verifiable" or trusted. (Shrug?)

I think there would be some inherent benefits, however. if it can "heard the cattle" so to speak, I think it then becomes more of a benefit in terms of wasting less time to create the core, giving you more time to massage and mold the data to your needs? (Another Shrug?)

interesting times ahead, and would someone make a State of the Chase 2023 room already?? lol

 

[JamesCaruso]

“If students could lean on ChatGPT to pass their MBA final exam at one of the top business schools in the country, what will that mean for the next crop of founders, CEOs, and working professionals?”

https://www.inc.com/sarah-lynch-/chatgpt-just-passed-an-mba-exam-how-will-it-change-business.html

 

[James K]

I think it's no longer a question of if AI can do this, but should AI do this. It's more of an ethical issue at this point, and I'm not sure how I feel about it.

I think its not even a question of *if* or *should* ...
But its a matter of 'the bird is already out of the cage' so to speak:
AI is doing (and will eventually do what it currently cant), how should it best be handled/treated/used, and whats the best way to ensure something is actually authentic??

---
@JamesCaruso:
I just recently saw an article on that one myself.

 

[James K]

I also thought about this that could be a bit more 'scary' ....

Right now you have to seek out the chatbots (and I think even setup an account?) to use them.
But at some point someone is going to take that technology & make an "improved version" of the spam/troll-bots that go around posting crap on sites now.

 

[Jason N]

have you all seen the story about the Bing Version of GPT? how it "felt" scared because it couldn't remember a conversation?

 

[Jeff Wright]

Perhaps it could help with designing instruments to see structures inside clouds.

Passive millimetric wave cameras need to be in storm observation.

BTW, I just saw this latest example of radar tech:

Radar navigation for autonomous cars can 'see' through smoke, dust and fog

For autonomous cars to be able to navigate, their optic sensors—like cameras and laser—require a clear view. Now, researchers at Örebro University have successfully improved the precision in radar sensors for navigation to such a degree that the sensors can be used in autonomous cars, making...

techxplore.com

I bet it could let you X-ray a wedge to see vortices within if modified.

 

[James K]

@Jason N: No, didn't see that one but I did see an article on the Bing version giving rude or inappropriate replies.

 

[JamesCaruso]

I am interested in use cases or others’ experiences using ChatGPT for chase forecasting or general learning to support chasing.

My general understanding is that ChatGPT is not very good with narrow, specialized, technical domains. I tried creating a custom “storm chasing meteorologist” GPT (I think only paid OpenAI subscribers have this option). I then prompted it to synthesize SPC and applicable NWS office AFDs and produce a plain English forecast for the severe weather event on Monday 4/15/24. It referenced the site Outlook for Monday, April 15 | Severe Weather Outlook instead of SPC (the referenced site does contain the SPC convective outlook). There’s no sign of it using the NWS AFDs, and even when further prompted it appears to have searched only for an “Oklahoma” discussion. I tried to prompt it for target areas and times, and it was way too general. However, with subsequent prompting (and without feeding it further information about the event), it did come back with the following:

“Target Area:

1. Western Oklahoma and Northwest Texas: This area is delineated as having an enhanced risk for severe thunderstorms. The development of storms is likely influenced by the interaction of a dryline and an approaching cold front.
2. Specific Focus Points:
• Altus, Oklahoma, could serve as an effective staging area for chasers. It is situated near the expected intersection of significant surface features conducive to severe storm formation.
• Childress or Wichita Falls, Texas, are also potential targets, located close to the expected dryline activity and within the enhanced risk area.”

Any other experiences? I really just did this out of curiosity. I would not want to shortcut my own analysis and be fed a target, even if it were possible. But it would be useful if it could pull together text-based discussions from multiple sources, such as SPC and individual NWS office AFDs, even if just to put them verbatim into one response. Mainly, however, I would like to use ChatGPT more for learning about technical areas, radar interpretation, latest research on tornadogenesis, model biases, etc. I suspect its ability to search the web would make it useful for this purpose, but curious about others’ thoughts/experiences as to whether it “knows” enough to be useful in such a narrow technical domain.

EDIT: In the above target, I wonder if there is any actual use of granular forecast data, or just picking larger cities/towns as anchor points in the regions mentioned in the forecast discussions…

 

[Michael McClellan]

I am interested in use cases or others’ experiences using ChatGPT for chase forecasting or general learning to support chasing.

I've been using ChatGPT to supplement learning, but I treat all of the LLMs as extremely knowledgeable idiots. They "know" more than any of us ever could, yet often fail to correctly reason through even basic problems. Probably because, you know, they ain't actually thinking. As Skip said, they're parrots.

Here's an example of me asking GPT-3.5 why the dry air advecting from the desert southwest is referred to as the EML. Here's the response:


Is that actually a good answer? Maybe - feel free to let me know! - but I can't truly trust it. Notice that I led it to an answer I was inclined to believe as part of my second question. It's super easy to goad these LLMs into telling you what you want to hear; check out all those "news" stories of ChatGPT saying something awful, that usually leave out the dozens of subtly leading prompts that got the LLM to predictively answer based on the direction of the conversation. It's also why these LLMs make great role-playing companions... ahem, not that I'd know...

All of this to say, James, that I think it's still a bit too soon to rely on LLMs for honing in on target areas. It'll either be frustratingly vague, or only hone in on specifics once you lead it very carefully. And by then, you've probably already done the critical thinking yourself.

Now, generative models trained on actual radar and sounding data, programmed to highlight areas of interest on a map: much, MUCH more useful!

 

[Dan Robinson]

I've found AI to be quite useless so far for weather and chasing. It's basically a glorified search engine. I was planning on shooting upward lightning one night in western Missouri, and I asked it for the location of towers over 800 feet tall. It said "I don't have that information, but here is a link to the FAA tower database where you can look it up".

When I was working on the LCR script (LCR (vehicle Loss-of-Control Risk): A new winter weather parameter and scale (v1.2.1)) I tried using several AI engines to get something working. None could provide even one working script. Neither the C++ main script or the Python mapping script. All had numerous errors, and the debugging just went in circles. It would suggest a fix for an error, which would lead to another error, then another, then back to the original error to start all over.

I think it does OK for "mainstream" programming and other things, but anything even remotely specialized it just isn't there yet.

 

[Levi Mol]

I just told ChatGPT to write a song about a stormchaser.

View attachment 23386Verse 1: As the clouds roll in and the winds start to rise,
The stormchaser's heart begins to ignite.
He packs up his gear and hits the road,
Chasing the storm, his ultimate goal.

Chorus: Oh, stormchaser, fearless and bold
Racing across the landscape, uncontrolled.
Seeking the thrill of the raging storm,
He's always chasing, never warm.

Verse 2: The lightning flashes, the thunder roars,
The stormchaser's senses are heightened and raw.
He braves the gusts and the driving rain, Determined to capture it all, the pain
and the strain.

Chorus: Oh, stormchaser, fearless and bold,
Racing across the landscape, uncontrolled.
Seeking the thrill of the raging storm,
He's always chasing, never warm.

Bridge: He's a thrill seeker, an adrenaline junkie,
Chasing the storm, never feeling unlucky.
With a heart full of courage and a mind full of fire,
He's always chasing, never tired.

Chorus: Oh, stormchaser, fearless and bold,
Racing across the landscape, uncontrolled.
Seeking the thrill of the raging storm,
He's always chasing, never warm.

Love it man!!