The only way that would be possible is if an llm were in a mode of always being trained and could count on being accurate. Which i dont think is possible? So tr

The only way that would be possible is if an llm were in a mode of always being trained and could count on being accurate. Which i dont think is possible? So trying to use the calculated weights as a massive compressed data source that knows everything is a mistake and is just leaning into the hype. which this is doing imo. The only thing it seems to enable is using natural language as an interface, instead of a programmed api. Text goes in, translates to something the computer understands, text comes back to the use. Instead of like rpc and and error code. And there is code auto complete.