GPT-3.5-turbo

  • Link: https://chat.openai.com

  • Family: GPT

  • Pretraining Architecture: Decoder

  • Pretraining Task: LM

  • Extension: ChatGPT takes a GPT3.5 (aka GPT3 Davinci-003) pretrained model and uses RLHF to finetune the model mostly like described in InstructGPT but with slight differences in the data collection. ChatGPT is also more than a model since it includes extensions for Memory Store and retrieval similar to BlenderBot3

  • Application: Dialog agents

  • Date (of first known publication): 10/2022

  • Num. Params: Same as GPT3

  • Corpus: Same as GPT3 + datasets generated for RLHF

  • License: Closed source, accessible through API

  • Lab: OpenAI

PreviousGPT-4NextPaLM 2

Last updated