How to cite text-davinci-003 (GPT-3.5) in academic paper?
Is it okay to cite the original paper of GPT-3 (Brown et al.) and InstructGPT (Ouyang et al.)?
Do they share same architecture?
text-davinci-003 is closed source, proprietary, and in its announcement we only get its features.
In model documentation, it is placed in the 3.5 category with -002 (maybe more marketing the performance), and that it is the first deployed using human preference training (RLHF).
| text-davinci-003 (Legacy) | Can do any language task with better quality, longer output, and consistent instruction-following than the curie, babbage, or ada models. Also supports some additional features such as inserting text. | 4,097 tokens | Up to Jun 2021 |
|---|---|---|---|
| text-davinci-002 (Legacy) | Similar capabilities to text-davinci-003 but trained with supervised fine-tuning instead of reinforcement learning | 4,097 tokens | Up to Jun 2021 |
It is also using a different tokenizer then at GPT-3 release, indicating a new model trained, but also a different tokenizer than gpt-3.5-turbo’s 100k.
"text-davinci-003": "p50k_base",
"text-davinci-002": "p50k_base",
"text-davinci-001": "r50k_base",
The citation of its existence could be merely “openai”.
1 Like
Hi and welcome to the Developer Forum!
A format similar to this seems to be fairly common
OpenAI. (Year of access). OpenAI GPT-3 API [text-davinci-003]. Available at: [URL of the API documentation or platform]
1 Like