>>512166211
yeah, by training I mean both the cost and the process itself
there's no thick brush strokes here, synthetic data can be extremely useful depending on the topic
there's so many interesting processes being forced by AI development right now
the AI-pipeline is "slowly" building itself, this implies data-collection, classification, curation, learning, and user-interaction
retards see LLMs as a whole engine, when in reality they're just one evolving piece
LLM, CoT, agents, now CoT is being replaced/integrated by hierarchical reasoning, and soon LLM models are going to be mini-agent systems that create a robust reasoning process with research models, validation modules, extrapolation modules, AI creativity hasn't even started yet