OpenAI Codex was built off MY work

Codex
1

I know this is going to come across as egotistical, and I don’t have any delusions on thinking I have any recourse, but I’m PISSED.

Codex is a direct ripoff of my ChatGPT plugin, Recombinant AI® that i built back in January of 2023 and released in July of the same year. Maybe this is dramatic but there is no doubt in my mind that my work contributed directly to the team that works on Codex.

Feel free to search these forums and you’ll find evidence. Recombinant AI® was my baby and I built it from scratch before Copilot could do anything but write annotations, or Codex or Cursor etc even existed.

I used Github’s API and the ChatGPT plugin framework, along with PluginLab with an IDENTICAL auth flow to connect people to public and their private Github repositories. The plugin would navigate HUGE codebases with multiple calls, traversing the tree, and could do exactly what Codex does, years ahead of it.

I made the FIRST conversational IDE, complete with a revolutionary but old school MS-DOS style, rendered UI, with external file storage, memory, custom instructions… everything. All before they were features on ChatGPT.

It worked exactly the same as Codex does now and id bet the framework isn’t that different under the hood.

Listen, Im trying not to have delusions of grandeur, and I’m honestly just venting… and maybe hoping to either be brought back down to earth, or validated, but I feel incredibly slighted that I built this so far ahead of the team here, but will never get any recognition or anything for it.

I even had a provisional patent, that I let lapse due to money, for a modular plugin system before OpenAI released Plugins/functions/tool use and id be lying if I did it doesn’t bother me.

I dunno… im just venting… but DAMN.

2

it was patented? as that’s what matters here…

You’d have had legal protection for your invention.

If Codex or similar tools used specific patented methods or systems, you could have pursued legal action or licensing agreements.

Recognition or compensation would be far more feasible because patent protection gives you enforceable rights.

3

what u mean Lmfao ANYOne can build a codex hahaah

they have the money man power and framework, they arent the only ones with coding systems

why not just make a better one if you have the skills to make the first version?

The image shows a terminal log with various INFO and WARNING messages, ending with a red error indicating "Error initializing ModelSelector: name 'BaseLLMProvider' is not defined. Using fallback model: gpt-4o". (Captioned by AI)
The image shows a terminal log with various INFO and WARNING messages, ending with a red error indicating "Error initializing ModelSelector: name 'BaseLLMProvider' is not defined. Using fallback model: gpt-4o". (Captioned by AI)756×196 8.1 KB

afterall… thats how us programmers get down

4

It WAS patented, and similar enough to the original plugin structure… but like I said I abandoned the patent because it would’ve cost me a couple thousand $ to complete and FAR more to do anything with… just wasnt in that position. This post is more directed toward how Codex is almost a direct copy of my ChatGPT plugin that let you talk to your repos holistically, and act as a semi- autonomous, cloud based, conversational IDE. I was one of the first 3 coding plugins on the plugin store. Check out my old posts. Just venting really lol

5

Not sure you fully read through my post.

7

Well I think that writing it at impulse does make it seem sort of like a need for validation, I do get to sense you tried to rectify it slightly (I hope it’s not used as a pseudo-alibi :innocent: ) when you mentioned to also be brought back down to earth, but reading your post worries me more about something else.

Putting aside your vent, if say, you’d have written this after cooling down (or I’d have helped you tone it down :sweat_smile: ), I am actually reading “your shared work is not safe, it can always become their property one way or another, they can use it either just as an idea and run with it, adapt or get inspired by your workflows, pipelines, jobs or processes, stamping it with their logo and publish it as if it was the newest, never before seen thing”.

Now granted, in the bigger scheme of all things, this can be a good thing as they do have the resources and manpower to improve anything more than we could (you said yourself, just patenting it would drain your wallet, let alone have a datacenter or other resources). And hey, that’s always a good thing.

… on the other hand, IF the fine print you laid out is true and not just a coincidence, then there’s a problem.

P.S. playing devil’s advocate a bit, trying to be neutral and hoping for good faith (although… seeing the scandals from behind the scenes - the way they took other people’s work without consent and using it to train their models being the big one, the political overview, the desire of dereglementation and more), we can say that a lot of similarities in workflow are like that because “it is the most optimal way to do it”. (hey, let me hope here)

I guess the best way to tell would be to see how many custom GPTs become official components / modules.

Still, regardless of it being “inspired” from your work or not, still sorry to hear about the outcome. I know how it feels to not get any recognition when you do 90% of the work (or 100 in your case) and someone else took and got credit.

8

I hear that you’re upset… and, tbh, this is such a fast moving field and everyone is copying everyone else. Software patents have never made sense (which is why they’re not even possible outside the US). Patents aren’t there to server some higher moral purpose, they exist to encourage more innovation to be shared. But there’s plenty of innovation in this field and it’s being shared pretty rapidly because it’s extremely competitive.

If your patent was somehow held valid, it would have slowed down innovation in this field and been a disservice to users of this technology worldwide. So I’m glad you didn’t try to actually patent and/or enforce this.

Also, I would gently puncture the view that “I came up with this”. It’s likely that there were dozens of people and teams coming up with that same idea all over the place, even at OpenAI.

As we say in the startup world, don’t be precious about your idea. If your idea is any good, there’s already 100+ people working on it around the world, 10 of them with more funding than you.

And I get that it’s annoying when it feels like someone steals “MY IDEA”… my advice is, let go. Ideas don’t belong to anyone. And we’re all better off as a society and as individuals when ideas can flow and evolve freely.

9

I would also let it go. It sucks, especially if you had a patent in the pipeline, but not finalized.

Just be glad that you were one of the first people to have figured this out.

11

@GlassAcres that’s a very uncomfortable feeling indeed.

I think it is important to share your original Topics to demonstrate just how long you’ve been working on this and the content of that work:

It’s a very rapidly evolving space with a lot of money and resources being applied so situations like this are bound to happen.

Monetising and protecting your work as a small/one person company is very challenging. I wonder what was in the T&C’s for plugin writers?

What I would do in your shoes?

*. I would consider open sourcing your plugin (if not already the case) and posting it as a repo. That way you have gain some value by association of the code and you might pick up some business that way.

*. Perhaps you could apply to join one of the teams working on a similar product at one of the well funded entities like OpenAI. Your experience would seem perfectly relevant? I imagine the salaries on offer might well exceed your plugin revenue and then some.

Best of luck!

12

Sorry you feel ripped off your baby. Not here to judge anyone or take any position, personally I saw so many ideas and realizations be done simultaneously and independently just because the context turned out to be favorable to do it, so can’t say more on that.

But what I’ve learned from my some 16+ years in development is this:

  1. Avoid patenting your software unless that’s the only way (don’t expose the details on how your thing works).
  2. If possible, make it work as a black box with approach “something in - unknown magic - cool stuff out”.
  3. Make it affordable (possibly as cheap as it can be compared to perceived value).
  4. Make it easier/more financially interesting to resell your software than to crack/reproduce/reverse-engineer it (affiliate, distribution channels etc ). This also applies to making it cheaper to buy your company than to compete with you.
  5. Once the #1-#4 are done, explode with communication so that you become the only guy that comes to one’s mind if they think about the problem you solve (even if it’s not their problem).

The 5 steps above is the only realistic way to protect your software I have found so far. If you think of anything else, I’ll take it.

Hope that helps.

13

There’s a lot that went into Codex, including fine-tuning o3. The overwhelming majority of their work is in the training, not auth flow and function schema.

Also, plugins were removed entirely quite a while ago. Unless your work was actively in use at the time OpenAI launched Codex, there’s zero legal recourse here, which I’m glad you acknowledge.

14

I pulled the “plugin terms” off archive.org, June 2023.

It isn’t the “we own every single word of your GPT and its files, to do anything we want forever” of today, but has:

  1. Indemnification; Disclaimer; Limitation of Liability

(a) Indemnity. You will defend, indemnify, and hold harmless us, our affiliates, and our personnel, from and against any third-party claims, losses, and expenses (including attorneys’ fees) arising from or relating to your Plugin, including Plugin Responses, your website or application that is connected to the Plugin, and your violation of these Terms or applicable law.

Plugins wasn’t a full set of instructions and files. It was an API connector. Think of it as “MCP v0.1 for ChatGPT”.

15

Just commenting…

I watch the industry as it evolves every day. There is a ton of duplication as many of us get similar ideas at the same time. This happens throughout history. That doesn’t cushion the blow when someone else publishes something that we’re passionate about before us, especially when it’s really close to what we’ve been working on.

Not to dismiss anything in this thread but to add to the conversation: Sam did say early on something like this:

Don’t assume OpenAI is missing something. If you do, you’ll probably be disappointed as the company continues to develop their own offerings. If you want to be successful, develop something completely different.

The “idea” of CLI assistants was with all of us from the start. Who can be a Linux CLI fan and resist the joy of using Custom GPT actions to open a CLI connection to process a ChatGPT prompt?

OpenAI just put a front-end on this. And for once I think they did a decent job with their first offering.
ChatGPT on the other hand is still an MVP in production. It’s horrible for so many reasons. We could all complain that OpenAI put a text box and submit button in front of a LLM, and that they let us save conversations in a left-hand nav-bar. That “idea” has never been well-developed by OpenAI (see all of the requests in Discord and elsewhere).

Again, if we compete with them on that, we’re likely to lose. But if many of us seriously re-imagine the chat bot with a lot of useful features, OpenAI won’t be able to keep up. We’ll all borrow ideas from one another as we all move the “platform” forward. I don’t think OpenAI or any of the other LLM providers will complain when we’re basing new offerings on “their idea” - of a texbox and a submit button in a browser, called a ChatBot.

That’s what I encourage you to do. Just move forward. See what they did well with Codex and do much better. Don’t assume they’ve missed something … do something really innovative and they probably won’t duplicate it. If they do … move forward. That’s the game.

HTH

16

I think LLMs are a great tool to help with your project or business, like by acting as an assistant or processing large amounts of simple tasks quickly, but simply wrapping a LLM and repackaging it is never going to be innovative. Because anyone can do that. And everyone already has.

Come up with something unique, build that, and then use AI to improve it. But don’t just dropship ChatGPT.

17

I hear you @GlassAcres - It must be shocking to see it come to this. I hope your visible history on the forum comforts you a bit. It’s traceable and I’m sure it will only strengthen your status and position as a pioneer in this field. What is next for you?

18

Trying to revolutionize building automation with various agentic structures to give stakeholders at every level accessible insights on how to optimize performance and energy efficiency.

All the big players like Seimens and Trane do it on a large scale for enterprise level data analysis. I’m trying to make cheaper, accessible AI driven systems that make it easy for anyone to keep their buildings and tenants happy without shelling out millions for something traditional machine learning does just fine already.

19

Oh i know that everything we all built back during the plugin phase was being harvested. Hell I chatted with the guy who built the first PDF Assistant plugin on the store. He made a nice chunk of change on it too.

Yeah plugins in general weren’t, but mine was :wink: I had a DB layer, 3rd party user management, and the github api. So basically you would sign up for my plugin and you’d get cloud storage associated with your user id. You could create projects and have custom instructions that were appended to your prompts (before chatgpt had those too), save project notes etc.

It meant you could have memory across multiple chats back in 2023 lol. Combine that with access to pull trees and files to Github, and push commits directly from ChatGPT and it was the first Conversational IDE. Basically what Replit’s Agent is now. I actually built and hosted it on Replit before that blew up to what it is now.

20

I gave up trying to keep up a while ago and just build stuff I like. It was just almost a blast from the past to open to Codex and be like..“wait…WTF… this is exactly like my plugin back in the day!” Lol

21

Parallel innovation is NUTS right? I’ll have to pull up my old patent for funsies, but I agree with everything you said. This post is honestly just a vent-y blast from the past sort of thing. I’m not actually upset it’s just a wild feeling.

I grew up with a guy who’s dad showed me some old drawings and expired patents he had put together from the 70s that were near identical to the first gen IPod… the tech just wasnt real yet :squinting_face_with_tongue:

22

If i were you i would just look for a new hobby.