Ideas to push the limits of ChatGPT

A woman holding up a painting of a lightbulb (Photo by Andrea Piacquadio)

ChatGPT’s role as assistant, proxy and human-like interface has been well marketed. It’s interesting to consider where the limits lie. Below are two questions I asked ChatGPT as part of a conversation documented in What does ChatGPT mean for CTOs?

Is ChatGPT able to participate in group discussions?

ChatGPT markets its conversational abilities. But, many human conversations involve more than one participant. Chatbots in business scenarios have traditionally attempted one-to-one conversation. Can ChatGPT go beyond this limitation?

A group conversation

Interviewer ❯
I have another, more general question about ChatGPT’s capabilities. Our discussion has revealed the strength of ChatGPT’s model in conducting a human like conversation. But, can ChatGPT successfully participate in group conversations?

ChatGPT ❯
ChatGPT is primarily designed to generate human-like text, and it has been trained on a large corpus of conversational data. In theory, it could be used to participate in group conversations, but it would likely require additional fine-tuning or customization to handle the complexity of group dynamics and different conversational styles.

One way to use ChatGPT in group conversations would be to have it generate responses based on the input it receives from the group. It could be trained to understand the context of the conversation and generate responses that are appropriate for the group.

Another way would be to use ChatGPT as a conversational agent that interacts with multiple users simultaneously, like in a chatbot scenario. However, this would require additional development and customization to handle multiple simultaneous conversations.

It’s worth noting that, as with any AI model, ChatGPT’s capabilities are determined by the data it was trained on, and the quality of the training data. It’s also important to consider that group conversations are more complex than one-to-one conversations, and it’s likely that ChatGPT would require additional fine-tuning and development to achieve a good performance in group conversations.

It’s not clear whether ChatGPT can succeed in group conversation. Much would depend on the depth of understanding in the model. This is a very interesting avenue for research. An AI could play a powerful role in influencing group thought. One can imagine many beneficial, and also nefarious, scenarios.

Is ChatGPT’s contextual learning portable between users / sessions?

So far, ChatGPT suggests simple conversation sessions. But, the contextual learning in a conversation could be valuable in other ways.

Consider, for example, an underwriting scenario. ChatGPT could converse with an insurance claimant, learning about the claimant’s context. Later, an underwriter and/or an investigator could use ChatGPT to simulate the claimant in conversation. ChatGPT might be able to function as a representative of other people.

Moving a conversation to a new context

Interviewer ❯
Considering the context which ChatGPT accumulates as part of conversations, is it possible to make the conversation and context available to another user as a starting point for discussion?

ChatGPT ❯
Yes, it is possible to make the conversation and context accumulated by ChatGPT available to another user as a starting point for discussion.

This can be achieved by storing the conversation logs, either in a database or a file, and making it accessible to other users. The conversation logs can be used to provide context to another user, who can then continue the conversation from where it left off.

Another way would be to save the context of the conversation, the model’s internal states, and the parameters of the model. This way, the next user can continue the conversation by providing the model with the previous context, and the model will be able to generate responses that are consistent with the previous conversation.

However, it’s important to note that the quality of the conversation will depend on the data the model was trained on, so it’s important to consider the quality of the conversation logs. Additionally, it’s important to consider the security, privacy and compliance aspects when handling and storing this kind of data.

It’s also worth noting that, as with any AI model, ChatGPT’s capabilities are determined by the data it was trained on, and the quality of the training data. So, it’s important to consider if the model has been trained on data that is similar to the use case you have in mind.

Interviewer ❯
Does the ChatGPT service currently offer this capability? For example, can I expose this conversation for others to use via a URL or API?

ChatGPT ❯
The ChatGPT service currently does not offer this capability as a built-in feature. However, as a developer, it’s possible to implement this functionality by storing the conversation logs and making them available to other users via a URL or API. This would require additional development and customization on your end.

You could use the API provided by OpenAI to generate the responses, and the store the conversation logs on your own backend. Once the conversation logs are stored, you could make them available to other users via a URL or API.

It’s important to keep in mind that this feature would require additional development and customization on your end, and that it’s important to consider the security, privacy and compliance aspects when handling and storing this kind of data. Additionally, you should also consider the legal implications of making these conversations available to other users, as you could be exposing sensitive information and you should ensure that you are in compliance with any data protection laws.

In summary, while the ChatGPT service itself doesn’t offer this capability out-of-the-box, it’s possible to implement this functionality by storing the conversation logs and making them available to other users via a URL or API.

Here then is another opportunity for investigation. Portable conversation contexts seem like a valuable business idea. Benefits (and potential concerns) could include:

1. asynchronous communication
2. simultaneous communication (by proxy) with large numbers of people.

It’s fun to dream up new uses of ChatGPT. As OpenAI clarifies the pricing model I’ll be watching to see what people dream up.