Don’t understand the cynicism either. Is this not way cooler than the latest pre-revenue series F marketing copy slop bot startup? To me this task looks less l

Don’t understand the cynicism either. Is this not way cooler than the latest pre-revenue series F marketing copy slop bot startup? To me this task looks less like next token prediction language modeling and more like translating a single “word” at a time into English. It’s a pretty tractable problem. The harder parts probably come from all the messiness of hearing and playing sounds underwater. I would imagine adapting to new vocab would be pretty clunky in an LLM based system. It would be interesting if it were able to add new words in real time.