Hi! My prompt has 12,635 characters, i’d like to know techniques and platforms to help me reduce the characters to reduce costs in GPT-4o assistant, thanks!
The cost is not measured in characters, but in tokens, the internal unit of AI language.
Tokenization of ideas is most efficient in English, because that is the largest part of the tokenization training dictionary. It acts as a compression about four characters to 1.
You may have overlapping reinforcement of ideas that is not needed in such a large prompt if it is all instruction. It depends on how much of that is the data to be operated on.
You can reduce your costs by using gpt-3.5-turbo-0125 -16k context, being significantly cheaper than gpt-4 models, or gpt-3.5-turbo-0613 if not using knowledge, where its 4k context will limit the maximum chat history and costs.