kalomaze@kalomazeah, so when a chinese company publishes their production codebases & models under permissive licenses, only then should you cry about foul play. really provocative technique they are going with hereOpens with an observation
kalomaze@kalomazeDeepSeek v3-0324 is a stable, solid, generalist model pretty straightforwardly, you can say this is an openly licensed "better than old sparse GPT4" model, w/ no reasoning attached. no doom loops or repetition spirals. decent in-context learning. improved je ne sais quoi...Opens with a number
kalomaze@kalomaze>extensively plan a coup so you can get the CEO fired over the course of a year >it never occurs to you to actually think about how the hundreds who are employed under his leadership might feel about this what did ilya mean by this?Opens with an observation