Dear OpenAI Team,
I am writing to share an idea for enhancing the customized ChatGPT API. As a regular user and a professional with 40 years of experience in geomatics, I have recognized the immense potential of your tool in my field. However, I believe an additional feature could make it even more valuable.
Main Idea: Integration and tracking of discussions in a customized API. Currently, each ChatGPT session is independent, limiting the continuity and depth of expertise the tool can offer. My suggestion is to allow a customized API to retain a history of discussions and topics addressed with a specific user. This would create a sort of “contextual memory” for the API, enabling it to build a dynamic knowledge base specific to the user.
I liken this to the image of a tree, where each discussion represents a branch which, although independent, connects to the central trunk of the API customization. This approach would allow for the accumulation of knowledge and experiences, progressively enriching the base, symbolizing the increasing depth and complexity of shared expertise.
I understand the privacy and data security challenges this might entail. However, I believe that with appropriate privacy settings, this feature could greatly enhance the personalization and utility of the API for users in technical and specialized fields.
I am convinced that this enhancement could be a significant step towards more advanced and personalized AI systems, offering even more targeted support and assistance to their users.
Thank you for taking the time to consider this suggestion. I hope it can inspire future innovations at OpenAI.
Sincerely,
Gilles Lavoie
With the new Assistants feature you can access a thread session but what your asking for would be more trouble than its worth, data poisoning is awful you would have to selectively clean the data and leave only the most relevant all of the time. If you want to save your session for each user and reference it later just ask the assistant to save the conversation, it will automatically save a file and give you the id so you can use it in the next session.
Just an FYI, GPT website chat is great for keeping track of conversation
Dear User,
I appreciate your concerns regarding data pollution and the need to maintain information relevance in an AI system. However, I want to emphasize that the situation is different with the use of a customized ChatGPT API, like the one I envision.
AI, particularly as developed by OpenAI, is designed to effectively handle and sort information, even when it’s complex and voluminous. In the scenario I described, where a single user interacts with the API, the issue of data pollution is significantly less problematic. This is because all interactions and information come from a single source: myself. Therefore, there’s less risk of contradictions or misinformation, as the data are consistent and centered around a specific area of expertise.
Regarding the Assistants feature to save and reference sessions, while useful, it doesn’t quite match what I envision. My aim is to create a more integrated and dynamic system, where the AI can learn and adapt continuously from our previous interactions, without the need to manually manage session files.
The GPT website chat platform is certainly an excellent tool for tracking conversations, but the idea here is to take this tracking and integration capability to a higher level, allowing for a smoother user experience and closer collaboration between the AI and myself.
Thank you for your feedback and suggestion. Your perspective is valuable and contributes to a broader reflection on the optimal use of AI.
Sincerely,
Gilles L.
1 Like
Yes AI in its current state is ultimately just an informational database. It fetches the data analyzes it and then summarizes what is important based on context and delivers a shortened or formatted version that is easier to understand (for both human and machine).
I wish it were that easy to just dismiss data poisoning that easy like its a non issue, but it is. The longer the conversation persists the worse it will become. The only viable way we currently have to avoid this is to delete irrelevant data and then retrain the model on the previous conversations. Vector databases are great but they still take meticulous configuration to get the desired results ( and will still have to be cleaned eventually)
Ultimately what your asking for is the AGI everyone is so hyped about. The system that not only does what AI does but also interacts with the necessary functions to accomplish the job. We ALL want the AI that can automatically clean it self properly and interact with the things we want to. They are getting closer but not quite there yet, we still have to manually filter conversations for the appropriate data.
FYI, you can already do this with your own database setup and the API. It’s what I do! You would use embeddings and RAG concepts to build the history for each person, dynamic to each instant interaction.
But realize it takes some gettin’ your hands dirty 