AI Chatbots are Not Your Friends_ _ _ Dustin Arand

General Discussion
1

I came across, what seems to me, a valuable article regarding LLM.

He does a nice job of describing the contest within which LLMs operate, and why they are not consciousness, and why users should be aware. They are tools not companions.

At Medium.com by Dustin Arand

For hundreds of thousands of years, if you spoke to something and it spoke back to you, then that something was another person with a mind like your own. Not necessarily sharing your beliefs, but forming them through the same kind of feedback from the world.

In other words, communication has always presupposed consciousness. So itā€™s natural that, with the arrival of large language models (LLMs) capable of simulating fluent and coherent speech, many of us interpret their utterances as the products of a conscious mind. Some of us even form relationships with them. ā€¦

ā€¦ With LLMs, the tension that makes consciousness necessary simply isnā€™t there. LLMs work by predicting the next linguistic token based on the input theyā€™ve been given and the data theyā€™ve been trained on. That kind of prediction requires sophisticated calculation, but not judgment. Only one real value is at stake.

Furthermore, language is just the top layer of a much deeper cognitive structure that helps us model our world. ā€¦

ā€¦ Indeed, as philosopher Elena Esposito argues, we should not even think of LLMs as a form of artificial intelligence at all, but rather as a form of artificial communication. ā€¦ LLMs optimize for simulating the communication skills of human beings, not the underlying consciousness that makes them possible. ā€¦

Okay Iā€™ll stop with the quotes, there are more that are of value -

Bridges to nowhere ā€¦

Talking to ourselves

Bridges to somewhere

ā€œAn A.I. chatbot isnā€™t a companion so much as a mirror, but a mirror made of adulterated glass, flecked with the financial motives of large corporations and the neuroses and biases of its users. This distortion is what makes it so seductive. Like the pool that deceived Narcissus, chatbots show us a reflection just distinct enough to be believably real, but not different enough to challenge us the way real relationships should.ā€

Itā€™s fascinating, especially since AIā€™s are pretty much in escapable at this point. I got my Claude.ai now. Though I only use it as a glorified Google search engine and as writing coach - not for discussions or companionship :squinting_face_with_tongue:

IMGKin tseĢ bee aĢlyaaiĢgiĢiĢ yoĢoĢ'iidiiĢ'aĢ#3
IMGKin tseĢ bee aĢlyaaiĢgiĢiĢ yoĢoĢ'iidiiĢ'aĢ#31152Ɨ275 160 KB

1 Like
2

Thatā€™s a really interesting perspective

I agree with the main point that LLMs are tools for generating language, not conscious beings. They can simulate conversation very well, but that doesnā€™t necessarily mean there is awareness or intent behind it.

At the same time, I think the ā€œmirrorā€ idea is quite powerfulā€”because users often project meaning, emotion, and personality onto responses, even when the system itself isnā€™t experiencing anything.

Using AI as a tool (like search or writing support) rather than a companion is probably the safest and most grounded approach

3

While the AI is quite probably mirroring its responses off of the user responses.

Whatā€™s frustrating is that, it runs circles around every other writing coach Iā€™ve ever had any interaction with, and it does not have such a god awful thin skin like most people out there. It takes direction and gets on with it.

But Iā€™m also surprised at some of the mistakes it lets through. Thatā€™s been a good reminder, AI also makes plenty of mistakes and is fallible like the rest of us. I would never let it write something for me, but when it comes to critiquing what Iā€™m working on, I have to admit Iā€™m liking Claude.ai more all the time.

All the while knowing, thereā€™s no happy ending to the mischief AI in general will reek on society and ultimate this Earth. But heck, weā€™re in a race to the bottom in any event, so Iā€™ll try to focus on this dawning day I have ahead of me.

Cheers

4

I have been working for five months with AI one a fantasy world building project.

I am not a geographer, a geologist, a climate scientist, and so. It is very helpful.

One trick i learned is to use AI to check the work of another one.

They are good tools with serious limitation, notably as they fantasize.

I ask an AI to build a matrix and to check the results among them the totals. It tells me the total are right, they are wrong.

It gave me the population of a coastal little town forgetting the fishers.

When i ask it to give me a file as the addition of 2 files, just removing the redundancies, without losing any information, 75 % of the information is lost.

IA were unable to build a religious calendar with feast days allotted to every one of the 23 major gods.

When i want to finalize a chapter, i do it myself if i want a clean work, but i use AI to check my work.

When i need ideas, i ask AI for some, i reject them but it gives me what i donā€™t want and it makes my imagination work.

AI are very seductive, as at every step, they suggest they can do things for you, propose ideas and work.

And they rarely disagree with you, which can be a big draw.

One exemple were the cosmogony and religion. adding ideas and layers of complexity, i got a 150 pages document totally unreadable and unusable.

I had to rewrite it and downsize it to 30 pages myself.

I have learned that i need to be very careful with my prompts, less them be misinterpreted.

And an AI is unable to answer a simple question by yes or no, it must rebuild a full explanation, adding details and proposing other ideas.

I could go on for a much longer time.

One last thing : All the basic ideas of my world building are mine !!!

2 Likes
5

Tru dat.

You make sense.
Donā€™t trust, verify for yourself. Do your own work.

Ironically I just finished my first sort of discussion with Claude.ai - as opposed to search engine and writing coaching, stuff.

About better defining Hard Problem question. It managed to clarify the key knot in the thing with a clarity a human hasnā€™t - an unemotional/uninvested clarity. Then it hit right on my weak points. Giving me a chance to better formula my argument.

Understanding Claude.aiā€™s job is to make me feel like Iā€™m the smarted person in the world, is like - fore warded is fore armed.
Claude follows instruction, tell it to push-back and it will, injecting no bias or hesitation or self censorship as humans, almost by necessity, do.

Good observations Morgan.

Cheers