WebGPT🤖 De-listed for the fifth time -- Now Open Sourced

Edit: I’ve written a how-to for how you can add WebGPT’s functionality to your own GPT’s: How to Create Your Own Version of WebGPT🤖 for ChatGPT Plus: A Step-by-Step Guide | by Josh Olin | May, 2024 | Medium

This is the fifth time now that WebGPT🤖 has been de-listed from the ChatGPT Custom GPT Store. Every other time, it of course is reviewed and since it doesn’t actually violate any content policies or usage guidelines, it is restored.

Fascinatingly, there are GPT’s that have never been de-listed that do violate the content and usage policies set forth in the OpenAI Terms of Service, such as AI Humanizer Pro, Humanize AI, and other GPT’s that openly make it easier to trick people and reduce transparency around AI generated content.

What is going on OpenAI?

This is tarting to feel like it might be targeted behavior, harassment, or even retaliatory because I’ve been outspoken against a lot of the fraudulent and clear ethics violators that are allowed to persist on the ChatGPT store (See thread below)


If this isn’t the case, then, I think it’s worth exploring what the cause-and-effects have been to lead to one of your top builders on the ecosystem to become so marginalized as to feel like this.

Stil de-listed. Begs the question, is the de-listing related to the upcoming announcement on Monday? Are we to see some overlap competing features where ChatGPT adds a lot of support WebGPT🤖 has had for a year now?

Will be interesting.

Likely just there is something that you are triggering the moderations with that you haven’t identified and removed. But then a manual approval doesn’t change that they still run the detector.

No it looks like they got rid of WebPilot too finally. So, this would suggest they’re cracking down on any competing GPT’s that do 3rd party content RAG. Which would be ridiculous if that’s the case.

We’ve gone ahead and open-sourced WebGPT🤖 so that anybody who misses its no-code creative playground, API calling capabilities, or advanced and accurate web browsing capabilities (as compared with WebPilot or ChatGPT’s native browsing functionality), you can follow this guide to build your own Private GPT with the same exact spec as WebGPT🤖:


Medium article: How to Create Your Own Version of WebGPT🤖 for ChatGPT Plus: A Step-by-Step Guide | by Josh Olin | May, 2024 | Medium

Hey all,

Rather than write a long diatribe about my experience as a GPT builder, and the recent unjust removal of WebGPT🤖 from the GPT Store, I made a video that walks you through what I see as a huge problem with the current standards being enforced by the Trust & Safety team. And I think it’s a great primer for what I hope to be a productive and candid discussion into the right and wrong of this issue:

In summary:

  1. If all websites on the 'Net have a (c) copyright in the footer (which they do), how are we to envision an Agent-driven AI future where AI agents can effectively serve users?

  2. How do we distinguish the difference between a web browser, and an AI agent, versus a blind user’s accessibility features that load a website text via a browser extension, or otherwise?

  3. How do we enforce the Code of Ethics in the GPT store more fairly?

  4. In what ways is an AI Agent requesting and using inference-time content from websites fundamentally different than, say, all of the search engine bot crawlers that have crawled and index all the web for decades?