Using ChatGPT to write comments/posts about dive equipment

[kimh]

Probably not news for most, but this article is worth adding to the thread. Author Eliezer Yudkowsky is one of the world's foremost AI specialists.

"Pausing AI Developments Isn't Enough. We Need to Shut it All Down"​

 

[kimh]

A couple more terminator divers, from Midjourney AI.

 

[undrwater]

A couple more terminator divers, from Midjourney AI.

View attachment 780033 View attachment 780034

They don't look very "swimmy", do they? Can you add some language that might do that?

 

[kimh]

They don't look very "swimmy", do they? Can you add some language that might do that?

It was surprisingly difficult to tweak the image in the direction I wanted, despite trying different approaches to the prompt. With current generative AI, it’s all about the prompts, of coutse- and the different systems interpret things differently, and each individual system is changing, too. These images are actually a couple weeks old, and my prompts might work differently if I tried them today. It’s all moving so fast.

 

[rob.mwpropane]

A couple more terminator divers, from Midjourney AI.

View attachment 780033 View attachment 780034

Geez.. can't wait to see that guy following me on a dive

 

[berndo]

I think all this AI stuff looks impressive at first glance and are fun to play around with. ChatGPT and these picture mash-up tools aren't AI though.
The chat tools give OK answers when you asked very very basic stuff on a topic they have lots and lots of sources on in the database. For the apple watch the tool just crawls all the marketing material that's out there (of which there is an eff ton) and mashes it together. Once you asked for specifics or stuff that's not super mainstream, you only get garbage out.

For the picture of the terminator, the tool just puts hoses and stuff on a terminator. It's just mashing pictures together. It's not AI. No artist would draw a picture like this. For instance, look at where the breathing hoses go, they just end, because the tool doesn't know what a hose is, it just seen hoses on pictures that are tagged as rebreather pictures.

I think this article about real AI is ridiculous, IMHO. None of the tools are actually remotely close to AI. It's essentially google search with a pretty good text generator tool.

When I hear people say they use these tools for 'research' it makes me cringe. Just read the wiki article on a given topic and you get much, much better results. The wiki articles have sources you can check and have been verified by other people.

 

[Angelo Farina]

I think there is a misconception here.
A chatbot is NOT AI.
AI is a program which mimic human brain operation, so it really understands and processes information using an adaptive neural network.
A chatbot is just a text manipulation tool running over a huge database of existing text.
No undrstanding is involved, it is just assembling new text rearranging the existing one.
For learning more, just ask to chatgpt how it works...
@berndo is right in his analysis.

 

[tursiops]

Many ScubaBoard posts are indistinguishable from something written by a ChatBot/AI system.

 

[berndo]

Many ScubaBoard posts are indistinguishable from something written by a ChatBot/AI system.

It's kind of the same problem. People read stuff from a huge database (the internet, which is full of inaccurate data) but don't have any real life experience or knowledge to distingish and judge what's bs and what's not and than produce text based upon the random stuff they read.
That also how we got all these conspiracies online these days.

 

[ginti]

I think there is a misconception here.
A chatbot is NOT AI.
AI is a program which mimic human brain operation, so it really understands and processes information using an adaptive neural network.
A chatbot is just a text manipulation tool running over a huge database of existing text.
No undrstanding is involved, it is just assembling new text rearranging the existing one.
For learning more, just ask to chatgpt how it works...
@berndo is right in his analysis.

Are you speaking of the chatbotS we find on social networks and other applications in general - or ChatGPT-based ones? Because ChatGPT is actually based on deep neural networks, according to Wikipedia (GPT-3 - Wikipedia) and the owners. Also, Wikipedia says that many text to image generators are based on neural networks (Text-to-image model - Wikipedia, DALL·E: Creating images from text).

Am I maybe missing your point? Or maybe you are generally speaking of "chatbots," not necessarily based on neural networks? (Because yes, there are many chatbots based on neural networks, sometimes quite sophisticated, but many are based on more trivial text manipulation techniques... Chatbot - Wikipedia)

I am asking here, given that I am definitely not an expert in the field

PS There are many definitions of AI - because, usually, different researchers have diverse views of what should and should not be considered part of it. However, most commonly, neural networks are just a subfield of machine learning, which in turn is a subfield of artificial intelligence - so AI is way more than just adaptive neural networks.

NOTE: I don't trust Wikipedia that much, but anyone can do a quick search to find out that these results are actually quite true (then I agree there is tons of stuff online that claims to be AI, but it is actually not AI). There are many references around, some even in the Wikipedia pages I linked.