ChatGPt and surveying

[mrodoc]

It cannot just be me who, last night, suddenly realised ChatGPT could probably draw, at the very least, adequate line surveys. So I sent it two scanned grubby pages from a note book and, after some questions, it put together a very adequate plan and elevation of a cave Tony Boycott and I surveyed a few years ago. I was suitably impressed. Might try it with closed loops next but I don't see it having a problem. Puts surveying back into the hands of non technically minded folks like me! Of course this will be leapfrogged by those who take their phones underground press a button and have it reel out on the screen immediately!

 

[andrew]

Definitely an area that is moving quickly. I tried about a year ago and got nothing useful out. It would be useful if you could share the notes and the output, so others can learn from it.

 

[Lankyman]

Stop now before it's too late! You're telling them where the Mendip Resistance Organisation (MRO) will be when the robot wars start.

 

[Logismos]

Coincidentally, several of the Cave Survey Group are together today at the TSG and we've been playing with AI a bit.

There's certainly some promise there, (if you're willing to overlook the ethical/environmental/social issues etc.) but we've found that there's still a constant issue with it hallucinating.
For example, when we had an AI look at some paper survey data it would look at the first few; assume that there's a trend; and then extrapolate the rest of the data out.

It's really good at at discrete tasks, where you can supervise it and validate it's output - stuff you may have written a little code for, or found an online tool. Eg. transforming a stations from A,B,C,D... to 1,2,3,4

 

[mrodoc]

You might be interested to see the raw data it used. I was a linear survey but I already had drawn it and it's the same. I just had to tell it scale etc.

 

[mrodoc]

Next time I will find out how it deals with closed loops.

 

[mrodoc]

I have the data from when Clive Westlake and I plus Rob Harper surveyed Wookey about 40 years ago but looking at it I think ChatGPT would request a neater transcript and I think it has been resurveyed anyway!

 

[andrew]

I have the data from when Clive Westlake and I plus Rob Harper surveyed Wookey about 40 years ago but looking at it I think ChatGPT would request a neater transcript and I think it has been resurveyed anyway!

All data welcome in the cave archive. Historic stuff is good. If it's in an archive it maybe useful later. Comparison is always interesting.

 

[mrodoc]

Better scan it and send it in - presumably MCRA

 

[andrew]

Better scan it and send it in - presumably MCRA

Survey data is a national archive https://www.cave-registry.org.uk/ with areas within. The web page is a bit broken, but the actual data bit is well maintained and robust. Easiest way for small amounts of data is send it to me or the person who manages the area you have data for, which in this case is also me.

You put up the original notes for the survey you tried, would you be willing to share the output as well?