When building with AI technology, you soon realize that you are prohibited from using lots of examples in a prompt due to the token limits of the large language model. Most large language models work on a token limit of 2048 which is equivalent to approximately 8,200 characters. With a more complex task to be completed, that might not be enough to prompt the AI to give you a relevant output. This is usually where you see users of AI make their first moves into the wonderful world of fine-tuning.
Fine-tuning can be daunting for those who do not code or do not have the technical knowledge to read through the documentation and send the relevant requests programmatically. To fine-tune the best AI large language model providers like AI21 or OpenAI, you'll need to structure your data in a very specific way which results in a JSONL file. If you don't have experience with JSON then this is another hurdle to overcome, it is no wonder that so many give up when they see all of these barriers to entry.
Fortunately, Riku is here to change the game and make fine-tuning more accessible to everyone. Fine-tuning has the major advantage of breaking free from the limits of tokens within a prompt allowing you to enter hundreds, thousands or even more examples to really get an exceptional output that you can't reach from prompt engineering. Fine-tuning also moves the tokens that you'd usually have in a prompt to the backend so when you run a fine-tuned model, you do it more efficiently and actually save costs.
