[kaveenk]

I've build a GPT3 and DALL-E2 integration for Discord. With it, you can ask GPT3 questions, have full conversations with GPT3 (just like ChatGPT), generate, vary, and redo images, and even optimize text prompts for image generation!

The conversation functionality also has medium term memory, and you can chat with it infinitely without having to worry about context/token limits. Moreover, in the next few days, I will be implementing long term and permanent memory for the bot using embeddings!

[Labo333]

Cool project!

I think ChatGPT is a different model from GPT-3, which you are using.

From https://openai.com/blog/chatgpt/:

> ChatGPT is fine-tuned from a model in the GPT-3.5 series

[kaveenk]

Thank you! You’re correct that ChatGPT is different from davinci. They’re both built on GPT3.5, but ChatGPT is trained on more conversational influences. However, davinci is really strong and will behave mostly like ChatGPT with the right conversation prompt :)

[DeWilde]

Yes, the GPT-3.5 model has been fine-tuned using RLHF, this is the text-davinci-003 you can use through OpenAI's API's.

Not sure if ChatGPT has some additional fine-tunings as you can get similar response using text-davinci-003, with the Chat prompt, and a temperature setting between 0.3 and 0.7.

[mycall]

Do you think GPT4 or GPT5, being multimodal, will do something similar?