I can't find a way, even with just the regular openai library.
#Is there a way to customize the bot's personality in python?
knotty rock
faint quarry
I can't find a way, even with just the regular openai library.
set the system message, in the system message you can set the AI's name and personality
knotty rock
thanks!
high mason
@faint quarry Is there a way to somehow personalize a model to some context? Like send it message once and remember it forever?
I'm thinking of sending it something like small form docs
faint quarry
set the system message, in the system message you can set the AI's name and pers...
@high mason ๐
high mason
How long does it remember that for though?
faint quarry
How long does it remember that for though?
Bro, you need to know it's different for each model! Some models have a 4k context size, and there are models with 32k, 128k, and even 200k. The amount it can remember is based on the model's context length.
faint quarry
Mhm, then I need to resend.
It's based on how you code it. You can set the system message to be fixed so that it sends with all the requests. This way, you don't have to send it as a user message. First, you need to learn more about how the OpenAI API works. Then, you will easily understand how it can be done.
high mason
That being said, I think I once saw an OpenAI feature where you could pay to train your own model or something like that.
high mason
It's based on how you code it. You can set the system message to be fixed so tha...
Mhm, but that equals more tokens right
high mason
That being said, I think I once saw an OpenAI feature where you could pay to tra...
If I need a bigger amount of context I'd need sth like this
faint quarry
Mhm, but that equals more tokens right
It's the same; even if you resend it as a user's message, it still uses tokens. It's even worse because you confuse the model about which part is the user's question and which is the context. As I said, first read and learn how the OpenAI Chat Completion API works, then you will understand how to do it.
high mason
That being said, I think I once saw an OpenAI feature where you could pay to tra...
Like you train it on a lot of context and it remembers forever
Like the chatbots on websites
faint quarry
Like you train it on a lot of context and it remembers forever
It requires a large dataset and costs to train, yet there's still no guarantee of success. Fine-tuning is aimed at making the model respond in the desired format, rather than specifically teaching it new information.
faint quarry
How do they do that
The same way perplexity.ai did it, I think it's just searching and collecting all the info online, using embedding to filter the specific info for answering the question, then using the filtered info as the model's context to answer the user's question.