6 ms·

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

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

BeefySwain 4y ago
Can you explain what you mean by "stealth model"? What is it, who discovered it and how, etc?
- qup 4y ago
  https://github.com/waylaidwanderer/node-chatgpt-api https://github.com/waylaidwanderer/node-chatgpt-api This repo talks about it.
  - TOMDM 4y ago
    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.
    - qup 4y ago
      > 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.
- btown 4y ago
  https://github.com/transitive-bullshit/chatgpt-api/blob/main/src/chatgpt-api.ts#L252-L279 https://github.com/transitive-bullshit/chatgpt-api/blob/main...
ElijahLynn 4y ago
They do have an official API here > https://openai.com/api/ https://openai.com/api/
- optimalsolver 4y ago
  That is not an API for ChatGPT.
  - ElijahLynn 4y ago
    Oh, I thought ChatGPT is based on Davinci, which is on their API page. Is ChatGPT not based on Davinci?
    - optimalsolver 4y ago
      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-human-preferences/ https://openai.com/blog/deep-reinforcement-learning-from-hum...
      - MattRix 4y ago
        It’s not a “very different model”, it’s still heavily based on davinci (aka GPT 3.5)
        geekrax 4y ago
        That's what makes it "different".
jeremycarter 4y ago
Great library! Thanks for sharing
braindead_in 4y ago
Is it possible to fine-tune it?
dfrey1 4y ago
Any updated Go libraries with it yet?

BeefySwain 4y ago

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

qup 4y ago
https://github.com/waylaidwanderer/node-chatgpt-api https://github.com/waylaidwanderer/node-chatgpt-api This repo talks about it.
- TOMDM 4y ago
  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.
  - qup 4y ago
    > 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.
btown 4y ago
https://github.com/transitive-bullshit/chatgpt-api/blob/main/src/chatgpt-api.ts#L252-L279 https://github.com/transitive-bullshit/chatgpt-api/blob/main...

ElijahLynn 4y ago

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

optimalsolver 4y ago
That is not an API for ChatGPT.
- ElijahLynn 4y ago
  Oh, I thought ChatGPT is based on Davinci, which is on their API page. Is ChatGPT not based on Davinci?
  - optimalsolver 4y ago
    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-human-preferences/ https://openai.com/blog/deep-reinforcement-learning-from-hum...
    - MattRix 4y ago
      It’s not a “very different model”, it’s still heavily based on davinci (aka GPT 3.5)
      - geekrax 4y ago
        That's what makes it "different".

jeremycarter 4y ago

Great library! Thanks for sharing

braindead_in 4y ago

Is it possible to fine-tune it?

dfrey1 4y ago

Any updated Go libraries with it yet?