Playing with ChatGPT AI

**incisor** · 26th January 2023, 12:03 PM

was playing with AI on the ChatGPT site this morning

is amazing what it comes up with

tried telling it where i wanted to go on the cooma trip and it came up with an interesting itinery

then i told it to make up a story about me going to the 75th in an old Landrover....

Dave had always been a fan of Land Rovers. So when he heard about the 75th anniversary celebration in Cooma, he knew he had to go. The only problem was that his only means of transportation was his trusty old Land Rover, which he had inherited from his grandfather. But Dave was determined to make the journey, and set out early one morning from Brisbane.
The first part of the trip went smoothly, but as he got further out of the city, the roads became more rugged and the terrain more challenging. Dave's Land Rover was put to the test as he navigated through rocky hills and muddy streams. But he pressed on, determined to make it to Cooma in time for the celebration.
As the day turned into night, Dave grew tired, but he didn't want to stop. He was so close to his destination and he couldn't wait to see all the other Land Rover enthusiasts who had made the journey. Finally, after hours of driving, he saw the lights of Cooma in the distance. He had made it!
Dave pulled into the campground, where the celebration was being held, and was greeted by a sea of Land Rovers. He couldn't believe how many people had come from all over the country to celebrate the 75th anniversary of this iconic vehicle. He spent the next two days talking to other Land Rover enthusiasts, sharing stories and tips, and admiring the other vehicles on display.
On the last day of the celebration, Dave took his Land Rover out for one last drive through the countryside before heading back home to Brisbane. It had been an unforgettable journey, and one that he would always treasure.
Years later, Dave passed down his trusty old Land Rover to his grandson, as his grandfather had done to him. And he could not wait to see his grandson, driving the same land rover, to the next Land Rover Anniversary, just like he did all those years ago.

gold... the technology is getting interesting

**Tins** · 26th January 2023, 12:25 PM

This guy has some interesting takes..

**V8Ian** · 26th January 2023, 06:17 PM

If Dave had used another piece of technology, GPS satnav or Google maps, he could have stayed on the sealed highways.

**V8Ian** · 26th January 2023, 06:20 PM

Who owns the copyright and responsibility for AI created prose?

**incisor** · 26th January 2023, 06:27 PM

Originally Posted by V8Ian

Who owns the copyright and responsibility for AI created prose?

Future wars have begun…

**Saitch** · 26th January 2023, 08:24 PM

Originally Posted by V8Ian

Who owns the copyright and responsibility for AI created prose?

At a stretch, Elon Gates!

**prelude** · 27th January 2023, 08:17 PM

The copyright stuff has already going on now for some of the AI picture services, search youtube. You can try mindjourney yourself for example. I tried "aulro" once, just to see what it would pick up and I ended up with a bunch of images that were not related at all (of course) but I did get some random letters in the picture here and there. I think this is because the system was able to identify those letters and randomly placed them in the picture, like it would a signature or something. I did not safe that picture unfortunately since it was just not interesting. Anyway, in other words; the system was trained with images from all over, including a great many with copyright on them. I think that using copyrighted imagery is perfectly fine, but when the computer "uses" parts of it, certainly when they are signatures, it can become problematic.

Funny thing is, where chatgpt understands you reasonably well, mindjourney does not. I asked it to make me a picture of a P38 range rover modified to be a camper and it just could not do it. Even when I asked it to just make me a "p38, off road, mud, rock crawling" I got this:

prelude_P38_range_rover_off_road_mud_rock_crawling_2bd01dd9-7aaf-4145-92e5-515cfebc5100.jpg

You can clearly see that it understands the concept of the P38 somewhat but it is also very clear that it takes hints from the classic, perhaps even more than hints. It is really a riot to play around with that thing for hours on end, I even took a subscription (for a while) on that service just to play with it some more

Having said that, back to the chatgpt AI. I have been playing around with it as well and I must admit that I have been testing it with edge cases on purpose every now and then. Generally speaking I would say it is a pretty darn clever trick. Ask it a concise question about a programming problem and it will write a piece of code for you that will work 8/10. It can even improve on issues which is very amazing. However, ask it questions say about cars or other "hot" topics and it is very clear that there is a left leaning bias in it's responses. Sure, when pushed you will be able to get a "balanced" response, for so far as possible but you do have to put in effort.

Example: I asked it about some car related stuff and although the information was mostly correct, it always ended with "but cars bad (environmentally speaking)". Pushing the AI and asking it specific downsides of EV's for counter arguments sake it would provide the information but it would not volunteer it, as it did about the downsides of ICE. I guess this could be due to a bias in the wider general public at this time and it was after all trained with "our" data OR it shows a bias in the researchers programming the damn thing.

In any case, it is a rather disturbing development. It will take, say, a decade or so for the AI's to be sufficiently reliable and ubiquitous (it's just to expensive to run for everyone at this time) to let it code anything of consequence on it's own but many IT jobs can certainly be dispensed with by then. Also, if you think that steering the public discourse around an ideology is seemingly easy these days with bots on twitter and what not, just wait for a little bit longer when chatgpt (got a BILLION dollars from microsoft) is the default interface in BING searches (and google is sure to follow) and you will be permanently policed by our AI overlords.

All in all, it is a great time to experiment right now so if you have the chance, play around with it! Since the costs are gargantuan at this time, the free beta will certainly not last and it will disappear behind a paywall or like I said behind a search engine making playing around more difficult. Just to give you an example of the cost: the entire chatgpt language model that is used at this time takes 1TB (1 terra byte) of memory, not disk space, that would be to slow, but actual RAM to run. A server with 1TB of ram is not cheap even these days and with millions of users microsoft (this thing purportedly runs in azure) surely won't keep this up forever...

-P

**muddy** · 27th January 2023, 08:31 PM

hi Dave sounds like a 60 th trip

**BradC** · 27th January 2023, 09:07 PM

Originally Posted by incisor

was playing with AI on the ChatGPT site this morning

is amazing what it comes up with

tried telling it where i wanted to go on the cooma trip and it came up with an interesting itinery

then i told it to make up a story about me going to the 75th in an old Landrover....

gold... the technology is getting interesting

Not much of a story, nary a spanner in sight!

**JDNSW** · 27th January 2023, 09:50 PM

As with most AI it is based on learning by looking at vast quantities of input or training data. There will be biases dependent on the selection of input data, but for this sort of AI it relies on the sheer volume to reduce biases - in any case, the bias tends to be what you could consider a consensus view.

Since a lot of the training data is actual scientific data, it is not surprising that some results come up to confirm the well known observation that "reality has a left wing bias". It doesn't actually - reality does not know about left/right. Just that some fringe elements seem to be good at denying reality. (Also called wishful thinking)

Thread: Playing with ChatGPT AI

Thread Tools

Playing with ChatGPT AI

Tags for this Thread

Bookmarks

Bookmarks

Posting Permissions