Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Can't wait for the official API.

In the meantime, we discovered a "stealth model" which is being used by some YC companies that ChatGPT uses under the hood. I just updated the chatgpt NPM package to use this stealth model w/ the official OpenAI completions API: https://github.com/transitive-bullshit/chatgpt-api



Can you explain what you mean by "stealth model"? What is it, who discovered it and how, etc?


https://github.com/waylaidwanderer/node-chatgpt-api

This repo talks about it.


This feels like a recipe for getting your OpenAI account banned, and probably fairly tbh.

I wonder if the fact that the API may expect user tokens, but this doesn't provide them may be correlated with the bug people are experiencing with getting other peoples completions.


> the bug people are experiencing with getting other peoples completions.

I hadn't heard about this, but I think I got one! I had a completion that really had me scratching my head.

Anyway I agree, if OpenAI is the kind of company to ban people, this is probably a good way to get banned.


https://github.com/transitive-bullshit/chatgpt-api/blob/main...


They do have an official API here > https://openai.com/api/


That is not an API for ChatGPT.


Oh, I thought ChatGPT is based on Davinci, which is on their API page. Is ChatGPT not based on Davinci?


ChatGPT has been retrained with a method called Reinforcement Learning with Human Advice [0], effectively making it a very different model:

[0] https://openai.com/blog/deep-reinforcement-learning-from-hum...


It’s not a “very different model”, it’s still heavily based on davinci (aka GPT 3.5)


That's what makes it "different".


Any updated Go libraries with it yet?


Great library! Thanks for sharing


Is it possible to fine-tune it?




Consider applying for YC's Winter 2026 batch! Applications are open till Nov 10

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: