I’ve been half listening to Davos talks while doing other stuff. I was barely attending to this discussion yet was suddenly grabbed by this breathtakingly implausible claim:
He claims that tokens per dollar per watt is now a key driver of GDP growth. He argues “the job of every economy and every firm in the economy is to translate these tokens into economic growth”. His optimism is grounded on the expectation that the cost of tokens per dollar per watt will rapidly reduce for, I assume, a number of reasons e.g. innovation in energy provision, more energy coming online, falling token prices. But… if LLMs are a mechanism to generate tokens at a near zero cost wouldn’t we assume that reduced unit costs just leads to the production of more tokens which requires more energy?
The more wastefully you use LLMs the more effective they tend to be. The best way to get them to work is to throw context at them. This suggests to me that even if there could be a linear relationship between token-availability and economic growth (which seems to tacitly assume all tasks are fundamentally informational and that LLMs can be deployed across all tasks) then greater availability is just going to produce greater demand. Am I missing something? I think it’s plausible that LLMs are becoming a factor in productivity in some fields but not always in a positive way. But the claim he’s make here just seems breathtakingly implausible.
Mark Carrigan 