For prototyping I'll just use off the shelf stuff, but I'm already thinking that the highest possible cost would be LLMs and multimodal stuff. I can use open source and/or fine-tuned models for each. If I have the use cases nailed down, I should be able to go smaller in LLM size, saving tons. Claude 3.5 and GPT-4o are like an Abrams tank. Incredibly powerful and capable but there are also other ways to cut down a tree that don't cost that much :D
Amazing post! Super intrigued on the costs to run the entire alftred at the end? between licences, storage and api calls.
Its crazy I've been picking up coding again thanks to AI :D its been 10 years since i last coded
I'd also love to know :D
For prototyping I'll just use off the shelf stuff, but I'm already thinking that the highest possible cost would be LLMs and multimodal stuff. I can use open source and/or fine-tuned models for each. If I have the use cases nailed down, I should be able to go smaller in LLM size, saving tons. Claude 3.5 and GPT-4o are like an Abrams tank. Incredibly powerful and capable but there are also other ways to cut down a tree that don't cost that much :D