AI Chatbots and their current limitations.

[an anarchist]

I have discovered AI chatbots this week and I think I have explored some limits it has.

Using ChatGPT and some versions of GPT4, my biggest gripe is that it cannot remember much. But supposedly the paid subscription to OpenAI's website which grants access to GPT4 will enable you to have longer conversations.

I have been discussing my theological framework with the AI, trying to identify points of weaknesses and strengths. And to see where it needs depth and how so. It takes a while for me to explain, then it remembers the entirety of it for a little while and I can go sort of in depth in the exploration of my beliefs. But then it will eventually forget the entire conversation once it's gone on too long and I have to restart from scratch

If an AI chatbot had the capacity to have an indefinite conversation, I think this would be so cool. And I bet that that is an eventuality.

I would like the AI to perhaps give more nuanced responses too.

Anyone else here have any experiences with AI chatbots? What do you feel are the limits of them currently?

 

[The Sum of Awe]

I have discovered AI chatbots this week and I think I have explored some limits it has.

Using ChatGPT and some versions of GPT4, my biggest gripe is that it cannot remember much. But supposedly the paid subscription to OpenAI's website which grants access to GPT4 will enable you to have longer conversations.

I have been discussing my theological framework with the AI, trying to identify points of weaknesses and strengths. And to see where it needs depth and how so. It takes a while for me to explain, then it remembers the entirety of it for a little while and I can go sort of in depth in the exploration of my beliefs. But then it will eventually forget the entire conversation once it's gone on too long and I have to restart from scratch

If an AI chatbot had the capacity to have an indefinite conversation, I think this would be so cool. And I bet that that is an eventuality.

I would like the AI to perhaps give more nuanced responses too.

Anyone else here have any experiences with AI chatbots? What do you feel are the limits of them currently?

The limits are they do not truly have interests, opinions, or feelings about the discussion at hand. They are not people, they are information software.

If you are trying to identify points of weaknesses and strengths, why not put all of this that you're typing into the ChatGPT box onto a Google Document and introspect the details yourself? Could make for an interesting book!

 

[an anarchist]

The limits are they do not truly have interests, opinions, or feelings about the discussion at hand. They are not people, they are information software.

This is something that my discussions with people on religion and philosophy will always have.
But is it really a limit?
Having no interests - I can make it solely focus on the refinement and exploration of my philosophy. It has much more literal material knowledge than me. And since it has no interests, it does not mind dedicating it's existence within the conversation to my theological understanding.
Opinions: it is perhaps truly unbiased, then.
Feelings: how is this relevant to the pursuit of truth? :0

If you are trying to identify points of weaknesses and strengths, why not put all of this that you're typing into the ChatGPT box onto a Google Document and introspect the details yourself? Could make for an interesting book!

I can do that. And I can put my thoughts online such as RF. And I can use the AI. I can do all these things simultaneously and each has their benefits.

 

[The Sum of Awe]

This is something that my discussions with people on religion and philosophy will always have.
But is it really a limit?
Having no interests - I can make it solely focus on the refinement and exploration of my philosophy. It has much more literal material knowledge than me. And since it has no interests, it does not mind dedicating it's existence within the conversation to my theological understanding.
Opinions: it is perhaps truly unbiased, then.
Feelings: how is this relevant to the pursuit of truth? :0

I can do that. And I can put my thoughts online such as RF. And I can use the AI. I can do all these things simultaneously and each has their benefits.

Fair enough, it was only a suggestion. You should do what works best for you.

Personally, a lot of my beliefs have emotional thought put into them, and beliefs are opinions, yes? So I think differing opinions, other than just facts, are important as well.

 

[Brickjectivity]

Its like an extremely complicated spelling program, except instead of spellings words it spells paragraphs and pages. This doesn't always work, because unlike spelling words there is always more than one way to write a page. The neural networks are statistical inferences about what humans might write based upon data that is available (usually from the web). They are spelling attempts rather than thinking attempts. Their creativity is actually the permitted statistical error in one variable or another. For example you can, as a data scientist, allow your AI program to have more wordy responses or more rhyming responses or responses that have to do with a subject...an so on.

 

[lewisnotmiller]

Its like an extremely complicated spelling program, except instead of spellings words it spells paragraphs and pages. This doesn't always work, because unlike spelling words there is always more than one way to write a page. The neural networks are statistical inferences about what humans might write based upon data that is available (usually from the web). They are spelling attempts rather than thinking attempts. Their creativity is actually the permitted statistical error in one variable or another. For example you can, as a data scientist, allow your AI program to have more wordy responses or more rhyming responses or responses that have to do with a subject...an so on.

Have you seen the new AI implementation in Khan Academy (khanmigo).

Pretty interesting example of what you're talking about here. The AI is adjusted to consider more options before responding, to more explicitly use different personas, to interpret mistakes rather than identify them, etc.

Pretty stunning imho, when you consider the limited time they've had to work with the AI.