Legit self-improving AGI?

[Wowo51]

I asked some questions of ChatGPT regarding self-improving agents. OpenAI has a clause in it's license that forbids the development of competing models with it's AI's output. It's all good. OpenAI did advertise that their AI beat 90% of the lawyers, but they've been pretty clear about ChatGPT not being qualified as a legal advisor as well. So would it be any type of a sound defense to cite ChatGPT if OpenAI ever becomes overaggressive legally on this issue? It would be amusing.

ChatGPT's legal opinion

User
I'm building an agent that is designed to solve complex problems by querying OpenAI's API many times. OpenAI's license forbids the creation of competing models with it's Output. Is my agent a competing model? How can I use OpenAI's Output when creating my agent? Can I self-improve my agent?

ChatGPT
Your project of building an agent designed to solve complex problems using OpenAI's API involves several important considerations under the current OpenAI use-case policy:

Definition of a Competing Model: A competing model generally refers to a system that replicates or aims to replace the core functionalities of OpenAI's models. If your agent is primarily a tool for problem-solving that leverages OpenAI's API for augmenting its capabilities, and not for replicating or replacing OpenAI's own models, it would typically not be considered a competing model.

Usage of OpenAI's Output: When using OpenAI's output, it's essential to adhere to OpenAI's use-case policy. This means:

Ensuring the outputs are not used to create a service that directly competes with OpenAI.
Not using the output to train other machine learning models without explicit permission, as this could potentially be seen as creating a derivative model.
Respecting the guidelines around content generation, such as not producing harmful or misleading information.
Self-improvement of Your Agent: Self-improvement of your agent using OpenAI's output can be a complex area. If by self-improvement you mean dynamically updating your agent's algorithms or responses based on the output from OpenAI's API, this could potentially be seen as training a new model with OpenAI's output, which generally requires explicit permission. It's important to differentiate between using OpenAI's API as a tool in your agent's problem-solving process and using the API to fundamentally train or improve the underlying mechanisms of your agent.

Best Practices:

Transparency: Clearly disclose the use of OpenAI's API in your agent.
Data Handling: Ensure that you handle data responsibly, in accordance with privacy laws and OpenAI's policies.
Regular Updates on Policies: Stay informed about any updates to OpenAI's use-case policy and terms of service.
Seeking Permission: If your use case falls into a gray area or if you plan to use the API output for training your model, it's advisable to seek explicit permission from OpenAI. They can provide guidance and clarify whether your intended use aligns with their policies.

It's crucial to review OpenAI's most current use-case policy and terms of service, as these guidelines can evolve. If you're unsure about any aspect of your project in relation to OpenAI's policies, the best course of action is to directly contact OpenAI for clarification and guidance.

User
I would like to improve my agent with OpenAI's API Output. My agent is not a model by your statement in 1. My agent does not incorporate a neural net or any machine learning in a traditional sense. It queries OpenAI's API many times to solve complex problems. My agent accepts a Prompt from the user. It analyzes the Prompt and attempts to solve it. It can analyze the state of progress obtained as a result of Output from OpenAI and try to achieve the objectives of the Prompt again. Is this type of agent a competing model?

ChatGPT
Based on your description, your agent does not appear to be a competing model in the traditional sense, as defined by OpenAI's policy. Your agent functions more as an intermediary or facilitator, using OpenAI's API to assist in problem-solving tasks. Here are some key points to consider:

Nature of the Agent: Since your agent does not incorporate a neural network or machine learning mechanisms, and instead relies on querying OpenAI's API to process and respond to user prompts, it is more of an application using the API rather than a competing model.

Improvement Process: The way your agent improves itself by analyzing the state of progress from OpenAI's output and attempting to achieve the objectives of the prompt again is crucial. This process seems to involve refining the query or understanding how to better use the API's output, rather than using the output to train or enhance a separate model. As long as the improvement does not involve the creation of a new model or the enhancement of an existing model using OpenAI's output, it generally falls within acceptable use.

Non-Competitive Nature: The agent does not seem to replicate or replace OpenAI's core functionalities but instead, leverages these functionalities to provide solutions to user prompts. This is a key aspect in determining whether a tool is competitive or complementary to OpenAI's offerings.

Adherence to Policies: As long as your use of the API adheres to OpenAI's policies regarding data handling, content generation, and use-case restrictions, your agent should be compliant with their terms of service.

Continuous Evaluation: It's important to regularly evaluate the functionalities and improvements of your agent to ensure they continue to comply with OpenAI's evolving policies.

Seeking Clarification if Needed: If there's any ambiguity or if your agent's functionalities evolve to a point where they might intersect with training or enhancing a machine learning model using OpenAI's output, it would be prudent to seek clarification or permission from OpenAI.

In summary, as long as your agent is using the API in a manner that aligns with OpenAI's current policies and is not attempting to replicate, replace, or train competing models with the API's output, it should not be considered a competing model. However, staying informed about and compliant with OpenAI's policies is essential, especially as your agent evolves.