Can't connect to Text-Davinci-003 model to chat

[north axle]

Does anyone experience the same thing where 11 April can't connect to Text-Davinci-003 model? My application worked fine at 07April, and I didn't touch my application since then.

[queen hull]

https://status.openai.com

[north axle]

Thanks for reply.seems like my internet is the problem.

[north axle]

Do you find GPT3.5 turbo response is a lot slower than text-davinci-003? It's disappointed, did I do something wrong? I have followed documents to change model version and other requirements.

[queen hull]

It can be slower, it's not your fault. Simply, sometimes the requests take some more time because of servers overloading.

[north axle]

Senpai, is it worth to build chatbot on text-davinci-003? I see not much different, including how much tokens it consumed. But the speed is the spotlight.

[queen hull]

Depends on your needs and budget

[north axle]

Price aside, all models do not have memory on previous chat. How did OpenAi do it to have memory of previous chat ? Would like to chat with you personally in discord if you are good with me.🥺

[queen hull]

You simply store the previous messages and give them back to the ai.

[north axle]

I read this method suggested previously, it is working but not efficient. OpenAi doesn't have built-in chat history checking?🥺

[queen hull]

No, they don't have an history feature. At least for now

[static sedge]

Hi there.
If we need to same history and give it back to openAi - for next message we will spend prev. count of tokens + tokens for new message. Right?

[north axle]

Yes, depends on how you code for the prompts content, it will pile up a lot tokens due to history chat. Have you experimented it? Would be happy if you could share the result.

[static sedge]

I have experiment with 4 strings and spend 250 tokens. I hope that sometimes we can use id from response to keep conversation

[north axle]

I don't get your idea, what's id here??

[static sedge]

When you ask OpenAI API you got conversation id in response. Maybe we can you this ID to set it as context. I mean use previous responses

[north axle]

how exactly you doing it? Feedback the previous response that application stored??? It would be helpful if you give one simple example.

[static sedge]

you can check example in the official documentation - https://platform.openai.com/docs/guides/chat/introduction
API call

`import openai

openai.ChatCompletion.create(
model="gpt-3.5-turbo",
messages=[
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Who won the world series in 2020?"},
{"role": "assistant", "content": "The Los Angeles Dodgers won the World Series in 2020."},
{"role": "user", "content": "Where was it played?"}
]
)`

response
{ 'id': 'chatcmpl-6p9XYPYSTTRi0xEviKjjilqrWU2Ve', 'object': 'chat.completion', 'created': 1677649420, 'model': 'gpt-3.5-turbo', 'usage': {'prompt_tokens': 56, 'completion_tokens': 31, 'total_tokens': 87}, 'choices': [ { 'message': { 'role': 'assistant', 'content': 'The 2020 World Series was played in Arlington, Texas at the Globe Life Field, which was the new home stadium for the Texas Rangers.'}, 'finish_reason': 'stop', 'index': 0 } ] }

I like to use this id - chatcmpl-6p9XYPYSTTRi0xEviKjjilqrWU2Ve , but I can't find how 😦

[north axle]

I see, I roughly got your idea. You mean this response is stored based on this unique id API created. So you assume there is a way to use this for next chat ?

But currently API didn't provide the function to use it yet. Maybe it's too overloaded with those responses, they don't even consider to store it temporarily in the server. So our application side has to store it and find a way to give back to API, that's why alot developers did it by feeding it back to API.No way to get around I guess.

[static sedge]

😦 So, I need to store all requests and responses to create something like chat.openai.com , right?

[queen hull]

Your can store them in a JSON file

With a generated ID you created randomnly