Just saw that MSFT loses $20/mo per user of GitHub Copilot.
It feels like LLMs are pretty dirt cheap for users right now and that won't last. I feel like I need to exploit all the value I can out of things like ChatGPT right now while it's cheap. They're gonna lock it down eventually and charge you crazy money. #llm #ai
@allenu Yeah, but by then we'll have models that can run locally.
@LouisIngenthron For me the benefit is that it’s run on data from the internet. Running it on my own code would be useful but only goes so far.
@allenu That'll still be true. You'll be able to download a training file with data from the internet. It'll probably even come with a service that has an auto-updater. The bigger the file, the more local resources it'll take, but the better the suggestions will be.
@LouisIngenthron ah gotcha. How big would the files be though? And who would be running the updates?
@allenu You can download LLM training models now that are anywhere from small (1-3GB) to huge (100s of GB). In theory, if you focused the training on one programming language or framework, you could modularize it and keep the sizes much smaller.
And I won't be surprised when there are competing services from various vendors offering these files.
