4 ms·

Does ChatGPT have an API yet?

shagie 4y ago
ChatGPT is an interface to GPT that maintains a history of previous statements. GPT has an API. For example: curl https://api.openai.com/v1/completions \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $OPENAI_API_KEY" \ -d '{ "model": "text-davinci-003", "prompt": "Marv is a chatbot that reluctantly answers questions with sarcastic responses:\n\nYou: How many pounds are in a kilogram?\nMarv: This again? There are 2.2 pounds in a kilogram. Please make a note of this.\nYou: What does HTML stand for?\nMarv: Was Google too busy? Hypertext Markup Language. The T is for try to ask better questions in the future.\nYou: When did the first airplane fly?\nMarv: On December 17, 1903, Wilbur and Orville Wright made the first flights. I wish they’d come and take me away.\nYou: What is the meaning of life?\nMarv: I’m not sure. I’ll ask my friend Google.\nYou: What time is it?\nMarv:", "temperature": 0.5, "max_tokens": 60, "top_p": 0.3, "frequency_penalty": 0.5, "presence_penalty": 0.0 }' If you continue to populate the prompt with the additional content, and question it will continue - note that this this will consume compute tokens at a rather impressive (and accelerating) rate.
- LelouBil 4y ago
  > ChatGPT is an interface to GPT that maintains a history of previous statements. Not exactly, ChatGPT was fine tuned on top of InstructGPT. > We trained this model using Reinforcement Learning from Human Feedback (RLHF), using the same methods as InstructGPT, but with slight differences in the data collection setup. From : https://openai.com/blog/chatgpt/ https://openai.com/blog/chatgpt/
- ffssffss 4y ago
  It's pretty affordable though: https://openai.com/api/pricing/ https://openai.com/api/pricing/ The most up to date LLM is $0.0200/1K tokens (I believe this is tokens returned, approximately 1 word = 1 token, but not quite). I set a billing limit of $50 a month and spent a couple hours playing around this afternoon. It has cost less than $1 so far.
  - shagie 4y ago
    There's also a "how much it reads to produce that data." If you go to https://beta.openai.com/account/usage https://beta.openai.com/account/usage you will see things like: text-davinci, 5 requests 187 prompt + 218 completion = 405 tokens Note at the bottom of https://beta.openai.com/playground/p/default-grammar?model=text-davinci-003 https://beta.openai.com/playground/p/default-grammar?model=t... - you will see a counter in the lower right corner. The default that this playground model opens with shows 15 tokens in the prompt and the friend chat ( https://beta.openai.com/playground/p/default-friend-chat?model=text-davinci-003 https://beta.openai.com/playground/p/default-friend-chat?mod... ) starts out with 28 tokens. Submitting it, and the conversation now has 41 tokens in it... and adding another line, 50 tokens... and the response, now 97 tokens. And so on.
    - ffssffss 4y ago
      I stand corrected. I'm sure they mentioned this at some point and I skipped over it.

ChatGPT is an interface to GPT that maintains a history of previous statements. GPT has an API. For example: curl https://api.openai.com/v1/completions \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $OPENAI_API_KEY" \ -d '{ "model": "text-davinci-003", "prompt": "Marv is a chatbot that reluctantly answers questions with sarcastic responses:\n\nYou: How many pounds are in a kilogram?\nMarv: This again? There are 2.2 pounds in a kilogram. Please make a note of this.\nYou: What does HTML stand for?\nMarv: Was Google too busy? Hypertext Markup Language. The T is for try to ask better questions in the future.\nYou: When did the first airplane fly?\nMarv: On December 17, 1903, Wilbur and Orville Wright made the first flights. I wish they’d come and take me away.\nYou: What is the meaning of life?\nMarv: I’m not sure. I’ll ask my friend Google.\nYou: What time is it?\nMarv:", "temperature": 0.5, "max_tokens": 60, "top_p": 0.3, "frequency_penalty": 0.5, "presence_penalty": 0.0 }' If you continue to populate the prompt with the additional content, and question it will continue - note that this this will consume compute tokens at a rather impressive (and accelerating) rate.

LelouBil 4y ago
> ChatGPT is an interface to GPT that maintains a history of previous statements. Not exactly, ChatGPT was fine tuned on top of InstructGPT. > We trained this model using Reinforcement Learning from Human Feedback (RLHF), using the same methods as InstructGPT, but with slight differences in the data collection setup. From : https://openai.com/blog/chatgpt/ https://openai.com/blog/chatgpt/
ffssffss 4y ago
It's pretty affordable though: https://openai.com/api/pricing/ https://openai.com/api/pricing/ The most up to date LLM is $0.0200/1K tokens (I believe this is tokens returned, approximately 1 word = 1 token, but not quite). I set a billing limit of $50 a month and spent a couple hours playing around this afternoon. It has cost less than $1 so far.
- shagie 4y ago
  There's also a "how much it reads to produce that data." If you go to https://beta.openai.com/account/usage https://beta.openai.com/account/usage you will see things like: text-davinci, 5 requests 187 prompt + 218 completion = 405 tokens Note at the bottom of https://beta.openai.com/playground/p/default-grammar?model=text-davinci-003 https://beta.openai.com/playground/p/default-grammar?model=t... - you will see a counter in the lower right corner. The default that this playground model opens with shows 15 tokens in the prompt and the friend chat ( https://beta.openai.com/playground/p/default-friend-chat?model=text-davinci-003 https://beta.openai.com/playground/p/default-friend-chat?mod... ) starts out with 28 tokens. Submitting it, and the conversation now has 41 tokens in it... and adding another line, 50 tokens... and the response, now 97 tokens. And so on.
  - ffssffss 4y ago
    I stand corrected. I'm sure they mentioned this at some point and I skipped over it.