I'm testing the different models for OpenAI, and I noticed that not all of them are developed or trained enough to give a reliable response.

The models I tested are the following:

model_engine = "text-davinci-003" model_engine = "davinci" model_engine = "curie" model_engine = "babbage" model_engine = "ada" 

I need to understand what the difference is between davinci and text-davinci-003, and how to improve the responses to match that response when you use ChatGPT.

2 Answers

I'll complement Rok Benko's answer with this great visual summary by Yao Fu <[email protected]>:

TL;DR

• text-davinci-003 is the newer and more capable model than davinci
• text-davinci-003 supports a longer context window than davinci
• text-davinci-003 was trained on a more recent dataset than davinci
• text-davinci-003 is cheaper than davinci
• text-davinci-003 is not available for fine-tuning, while davinci is

Capabilities

As stated in the official OpenAI article:

While both davinci and text-davinci-003 are powerful models, they differ in a few key ways.

text-davinci-003 is the newer and more capable model, designed specifically for instruction-following tasks. This enables it to respond concisely and more accurately - even in zero-shot scenarios, i.e. without the need for any examples given in the prompt. davinci, on the other hand, can be fine-tuned on a specific task, which can make it very effective if you have access to at least a few hundred training examples.

Additionally, text-davinci-003 supports a longer context window (max prompt+completion length) than davinci - 4097 tokens compared to davinci's 2049.

Finally, text-davinci-003 was trained on a more recent dataset, containing data up to June 2021. These updates, along with its support for Inserting text, make text-davinci-003 a particularly versatile and powerful model we recommend for most use-cases.

Use text-davinci-003 because the other models you mentioned in your question are less capable.

If you buy a ChatGPT Plus subscription, you can also use gpt-3.5-turbo or gpt-4. So, to get similar responses as you get from ChatGPT, it depends on whether you are subscribed or not. For sure, gpt-3.5-turbo and gpt-4 are even more capable than text-davinci-003.

Costs

text-davinci-003 is cheaper than davinci, as stated on the official OpenAI website:

Fine-tuning availability

text-davinci-003 is not available for fine-tuning, while davinci is, as stated in the official OpenAI documentation:

Fine-tuning is currently only available for the following base models: davinci, curie, babbage, and ada. These are the original models that do not have any instruction following training (like text-davinci-003 does for example).

ncG1vNJzZmirpJawrLvVnqmfpJ%2Bse6S7zGiorp2jqbawutJobm9pZ2iBcoCOqKeeppGeeqi802ZqZpmgnnq4tMCtZKKrXam1pnnDop2fnaKau6SxjJucra%2BVmrtusMCvoKebmWKur7CMrZyxrF2Zrre1zZygZmhgaA%3D%3D