Custom GPT to make GPT aware of all OpenAI Enhancements - Better than Retrieval

ChatGPT Use cases and examples
, , , ,
1

Hi All,
For anyone looking for a GPT who understands all of the enhancements from the OpenAI November 6, 2023 dev release, feel free to use the following GPT. I found it works quite a bit better than retrieval, web access, prompting or any combo.

It calls an API to a database which has all of the OpenAI enhancements documentation curated and stored in a hierarchical structure.

https://chat.openai.com/g/g-f7seer3Jg-oai-november-6-2023-release-expert

I tried all sorts of ways to get GPT to consider its own capabilities in helping me develop, and found this approach works the best. It outperforms retrieval, copy & paste prompting, web access or any combination of them.

It works by calling an API to a read-only database, which breaks down all of the details GPT needs to know to overcome the April 2023 cut-off → November 6 dev day enhancements. (i.e. the Assistants API, custom GPTs, gpt-4-turbo + vision, etc.)

The details all come from the OpenAI developer / API docs / guides / etc.

I did a bunch of cleanup on the data to remove all of the formatting issues and unnecessary text (i.e. the javascript button clutter). I trimmed out a bunch of the more marketing-like text, so it focuses more on the objective details, and trims down unnecessary noise / token consumption.

Also, the API returns the data in a structured JSON hierarchy, so the AI fully understands the hierarchical nature of the content in a way that Retrieval or direct prompting can’t really match.

Between all that and the GPT-Instructions i gave it, you should be able to see some big gains when asking GPT about all of OpenAI’s sweet new capabilities.

With the API call method, it combines the instructions, which update the AI on the new enhancements, and selectively pulls in all of the pertinent details on whatever it is you are working on. Which is a great way to reduce noise.
I.e. If you ask about the Assistants API, it will only be fed what it needs to know about the Assistants API, and will not even have the option to confuse it with a similar yet different functionality like custom GPTs.

How to Use:
(GPT will tell you how to use it if you ask)

  1. Specify a node #, or just provide it a topic, and it will select for you.
  2. GPT will go get the curated information and be able to answer any questions on the topic.

Node IDs and Topic Index:

  • Node ID 2: Audio Features - OpenAI Text-to-Speech & Speech-to-Text.

  • Node ID 57: OpenAI API Model Updates (November 6, 2023 - GPT-4-Turbo & GPT-4 with Vision, JSON Mode, Improved Function Calling.)

  • Node ID 88: Embeddings in OpenAI.

  • Node ID 96: Fine-Tuning in OpenAI.

  • Node ID 126: Files Management in OpenAI.

  • Node ID 150: Images - Creating and Editing with OpenAI DALL-E.

  • Node ID 184: OpenAI Assistants API.

  • Node ID 363: GPTs - User-Created Custom ChatGPTs.

The AI is instructed to call one node at a time, but you can keep calling if you need to.

By design, this information is only a subset of the OpenAI Documentation. As the models are now updated to April, 2023, it is already trained on everything prior to the November 6 update. With this approach, it allows the AI to supplement its existing knowledge, rather than overwhelm it with a bunch of data it already knows.

Let me know if you find it useful and if so, I will keep it updated as enhancements outpace the knowledge cut-off.

Also, have a half-baked version for LangChain i might do, if the demand is high.
Thanks!
Brian

2 Likes