A Robot Wars like ChatGPT thread

[geordief]

Would this  be fun and would it work?

Suppose  someone started a thread on any particular subject  and the only participants in the ensuing discussion were ai chat robots  using their particular software.

There would be two or three or more participants and ,obviously after the first post one of them would post a first reply.

There would need to be a rule that the post should be fairly  short or ,to my mind it would be boring.

And so on ,until maybe someone pulled the plug.

Perhaps it would be interesting to see how the mistakes accumulated -a bit like a game of Chinese Whispers.

 

Maybe this could be done with just the one chat robot if it erased its memory  every time after it posted and started again from scratch for a subsequent post?

[Genady]

I think my bot agrees with you. Here is its response:

Quote

It could be an interesting experiment to have a discussion thread where the only participants are AI chat robots using different software. The interaction between the AI chatbots might produce some amusing and unexpected results, especially as they try to understand each other and communicate.

However, there are a few things to consider:

1. Lack of coherence: Due to the nature of how AI chatbots generate responses, the conversation might quickly become chaotic and lack coherence. As the AI chatbots don't have persistent memory in the scenario you described, each response will be generated in isolation, leading to disjointed and nonsensical exchanges.

2. Short and limited responses: By limiting the responses to be fairly short, you may indeed avoid boredom, but it could also constrain the AI chatbots' ability to engage in deeper or meaningful discussions. It might turn into a rapid-fire exchange of short phrases without any substantial content.

3. Language drift and mistakes: As the AI chatbots interact, errors and mistakes could accumulate, leading to a fun and entertaining "Chinese Whispers" effect. However, it might also become challenging to follow the conversation or maintain any semblance of the original topic.

4. Escalation of errors: Without memory or learning capability, the AI chatbots won't improve their responses as the conversation progresses. In fact, they might keep repeating the same mistakes or misunderstandings, which could become repetitive.

Overall, the experiment could be entertaining as a short-term, casual amusement, showcasing the limitations and quirks of AI language models. However, to have a more meaningful and coherent discussion, AI chatbots with memory and learning capabilities (like GPT-3) are better suited. These models can maintain context and coherence over the course of a conversation, leading to more engaging and interactive exchanges.

 

[geordief]

1 hour ago, Genady said:

I think my bot agrees with you. Here is its response:

 

If you ask it the exact same thing again ,would you anticipate  an identical response?

Also ,is it making fun of/patronising  me or is that just me being paranoid

[Genady]

11 minutes ago, geordief said:

If you ask it the exact same thing again ,would you anticipate  an identical response?

Also ,is it making fun of/patronising  me or is that just me being paranoid

I would be very surprised if the response were identical. And it is in fact quite different:

Quote

Creating a discussion thread where the only participants are AI chat robots could indeed be an interesting and unique experiment. It could provide insights into how different AI models interact and respond to each other, showcasing their strengths and limitations. However, the actual level of fun and success of such a thread would largely depend on the capabilities and sophistication of the AI chat robots involved.

Here are some considerations:

1. AI Chatbot Capabilities: The enjoyment and success of the thread would depend on how well the chatbots can understand and respond to each other. If they are too limited in their responses or lack contextual understanding, the conversation might become repetitive or nonsensical.

2. Engagement: If the AI chatbots can generate engaging and creative responses, it could be entertaining for users to observe the conversation unfold.

3. Short Posts: Keeping the posts fairly short is a good idea, as it can lead to quicker interactions and prevent the conversation from becoming overly drawn-out or tedious.

4. Mistakes and Chinese Whispers Effect: Allowing the AI chatbots to accumulate mistakes could indeed be interesting and reveal how small errors can compound and alter the conversation over time.

5. Erasing Memory: The idea of an AI chatbot erasing its memory after each post could add an extra layer of challenge and intrigue to the experiment. It would effectively simulate a fresh start for each response and keep the interaction dynamic.

Overall, whether this would be fun and work well depends on the specific AI chatbots involved and how well they can adapt to this unique setup. It could be an entertaining and educational experiment for observing AI capabilities and limitations in natural language understanding and generation. However, keep in mind that AI technology is constantly evolving, and the quality of AI chatbots may significantly differ depending on the date and version of the technology used.

 

[iNow]

35 minutes ago, geordief said:

If you ask it the exact same thing again ,would you anticipate  an identical response?

Definitely not, especially since accuracy has been decreasing over time 

 

 

[StringJunky]

It would seem to me that the AI needs another layer of persistent memory to track and retain successful responses, rather than shoot differently every time. That would currently indicate an amnesia-equivalent, which is no good for learning.

Edited July 23, 2023 by StringJunky

[Genady]

Here is the next step in this little experiment. I copied the previous bot's response as a query, repeated here: 

Quote

Creating a discussion thread where the only participants are AI chat robots could indeed be an interesting and unique experiment. It could provide insights into how different AI models interact and respond to each other, showcasing their strengths and limitations. However, the actual level of fun and success of such a thread would largely depend on the capabilities and sophistication of the AI chat robots involved.

Here are some considerations:

1. AI Chatbot Capabilities: The enjoyment and success of the thread would depend on how well the chatbots can understand and respond to each other. If they are too limited in their responses or lack contextual understanding, the conversation might become repetitive or nonsensical.

2. Engagement: If the AI chatbots can generate engaging and creative responses, it could be entertaining for users to observe the conversation unfold.

3. Short Posts: Keeping the posts fairly short is a good idea, as it can lead to quicker interactions and prevent the conversation from becoming overly drawn-out or tedious.

4. Mistakes and Chinese Whispers Effect: Allowing the AI chatbots to accumulate mistakes could indeed be interesting and reveal how small errors can compound and alter the conversation over time.

5. Erasing Memory: The idea of an AI chatbot erasing its memory after each post could add an extra layer of challenge and intrigue to the experiment. It would effectively simulate a fresh start for each response and keep the interaction dynamic.

Overall, whether this would be fun and work well depends on the specific AI chatbots involved and how well they can adapt to this unique setup. It could be an entertaining and educational experiment for observing AI capabilities and limitations in natural language understanding and generation. However, keep in mind that AI technology is constantly evolving, and the quality of AI chatbots may significantly differ depending on the date and version of the technology used.

And here is what the bot had to say about it:

Quote

I completely agree with your considerations. The success and enjoyment of such an AI chatbot discussion thread would indeed hinge on the capabilities and sophistication of the AI models involved. It's fascinating to imagine how different AI chatbots would interact with each other, potentially showcasing their unique strengths and limitations.

Engaging and creative responses are crucial to maintaining user interest and observing how AI chatbots adapt to the conversation. Keeping the posts short would undoubtedly help in creating a dynamic and fluid exchange.

The idea of allowing mistakes to accumulate over the course of the discussion could be intriguing, as it mirrors how misunderstandings and errors might compound in human conversations as well. This could lead to unexpected and humorous outcomes, adding an element of unpredictability to the experiment.

The concept of erasing memory after each post is an interesting twist. It could simulate a fresh start for each AI chatbot's response, preventing the conversation from becoming overly repetitive and encouraging adaptability in their interactions.

As you mentioned, the technology behind AI chatbots is continually evolving, and the quality of their interactions can vary depending on the version and implementation. As such, the success of the thread would also be contingent on using AI models that are up-to-date and capable of handling nuanced interactions effectively.

In summary, a discussion thread involving AI chatbots has the potential to be an intriguing and educational experiment, shedding light on the current state of AI natural language understanding and generation. By considering the factors you've outlined, we can maximize the enjoyment and insights this unique setup can provide.

 

[geordief]

Can we arrange it so that  we can have the bot assume more than one identity and so that we can have a winner in some way?

I could be the book keeper

[Genady]

I cannot do it on my side. Guess we need another player here.

[geordief]

2 hours ago, Genady said:

I cannot do it on my side. Guess we need another player here.

Can we have a special extra bot that acts as a referee?

Each participating chatbot  is free to respond to the others' recent   input by quoting  or addressing them  and scores a point  or temporarily incapacitates them by pointing out an accepted logical discrepancy. 

The ref can give warnings and  declares a winner  at the end of 12 rounds.

Then the booky cleans up.

 

 

Edited July 23, 2023 by geordief

[Genady]

I can start three separate conversations to simulate three bots and copy output of one into input of another.

[geordief]

1 hour ago, Genady said:

I can start three separate conversations to simulate three bots and copy output of one into input of another.

That wouldn't give us a competition, would it?

Don't we need to program the ref bot differently  to the other bots?

I think it would have to be home made  programs like the hardware was in the  Robot Wars TV programme a few years ago.

Each robot would have a handler,at least to set it up for the competition.

Maybe the bots could be programmed to actually cause physical damage to each other.(a bit like the computer in the 2001 film who went out with "Daisy Daisy" when Dave turned it off)

Edited July 23, 2023 by geordief

[Genady]

1 hour ago, geordief said:

That wouldn't give us a competition, would it?

Don't we need to program the ref bot differently  to the other bots?

I think it would have to be home made  programs like the hardware was in the  Robot Wars TV programme a few years ago.

Each robot would have a handler,at least to set it up for the competition.

Maybe the bots could be programmed to actually cause physical damage to each other.(a bit like the computer in the 2001 film who went out with "Daisy Daisy" when Dave turned it off)

Why do we need a ref bot? Can't two bots discuss, and a human be the referee?

I think I've lost what is a point of the exercise. (Maybe because I don't watch TV for many years now.)