Question to ponder: Is inconsistent or infrequent specific testing worse than not testing at all for monitoring training impulse response? Follow up question: Should we be monitoring median median power rather than mean max power to look for training impulse response?

I absolutely love that these things look all wrong

this seems bad

it’s a bit early in the year for this

Trying some rapid prototyping:

whats the over under that it works/ is usable?

Look if Microsoft Copilot can’t walk you through fixing a corrupt file without putting your computer into a spinning ring of death requiring a full windows reinstall then uh maybe it’s not ready for prime time.

This fuxkinf thing how are these LLM companies that are claiming they will replace 2/3 of the worlds work force getting away with launching pre-alpha apps that fail at the most basic reliability/functionality?

veloclinic.com/can-claude-rep…