Assistants API Settings Retrieval and Temp

I am using the assistants API but I don’t know how deterministic it is ( temp )

is the temp all the way to 0.1 or is it 0.5 ( average )

I would like to view the settings, where can I do so

also for the retrieval, how is it chunking the data, I would also like to see / adjust

How can I do so

The temperature is not in a user’s control and and not told to us. It is high enough to be an issue of reliability, so I’m thinking the defaults of 1, although there could be other methods going on for dynamically adapting sampling parameters and thus why control can’t be turned over.

The retrieval chunking method is undisclosed. It is only employed in a function that AI can use to search, thus returning chunks of information, if such a search can even be called “chunked” instead of “indexed”.

You can’t make the AI write at the length you’d want in order to read back what is placed into context, vs the 100k of input that is possible at $1 an experiment, the AI is reluctant to write beyond 800 tokens or so. One could go at the problem sideways.

What is also injected into the AI context to fill it is very naive.

The better solution is to never send assistants a token. “hello” will be multiplied by thousands in cost. Use chat completions.

wow that’s a very nice response,

I wanna have a “custom chatbot”

Do I build my own rag? I thought Assistants API ( gpt 4 model ) just charges like GPT 4 API, plus only the 0.2$/GB