Gpt-3.5-turbo-instruct is NOT as good as Davinci-003!

I really dont understand why they want to remove davinci when its so much better than any of the chatGPT models for creating articles. The newer models might be better at other stuff, but for generating text davinci is by far better with a good prompt. It nicely follows instructions, it creates creative text while the chat models often ignore instructions, look more generic and lack the same creativity.

To the OpenAI dev’s: Is there no way to keep davinci-003?

I don’t see what the problem is to just keep it around for people who prefer it.


Hello @styx

Can you share any examples that show the difference between both models?

It is a drop-in, merely in that it is also a completion model. 1/10th the price also tells you they found equal proportion of optimization.

Given the wide difference of 10 months of “instruct”, change is expected.

Identically prefixed and multishot prompt, top_p and temperature = 0.0


AI: Depending on the type of AI model you are changing, you could experience changes in accuracy, speed, scalability, and cost. Additionally, you may experience changes in the types of tasks the AI model can perform, as well as changes in the types of data the AI model can process.


AI: There are a few potential changes that could occur when changing AI models. One is the accuracy of the model, as different models may have varying levels of accuracy for different tasks. Another change could be the speed of the model, as some models may be faster or slower than others. Additionally, the complexity and interpretability of the model may also differ, which could impact how easily it can be understood and modified. It’s important to carefully consider these factors when choosing an AI model for your specific needs.

Glad I found people who are like me in preferring to use davinci and text-davinci-003 models instead of gpt-[34] for some tasks!

I have not used a lot yet, but model davinci-002 seems promising (not talking about text-davinci-002) …