@alexolegimas Taking data at face value, question is if slower adoption of Fable is evidence we are locally saturating marginal demand for higher intelligence / capabilities. Given cost/fixed unit of intelligence is declining fast, need Q demanded at the frontier to keep growing very rapidly
Many reasons why this wouldn’t be the case (RSI, limits of distillation, switching costs, etc) - but it is conceivable that a fully commoditized model layer wherein impossible to recoup training costs in economic profit would be bad for semis/world.
@duncancampbell@johnarnold Related point, but much like stocks I think people index on their expectations for the future moreso vs. present conditions. To extent those expectations have diminished / become more uncertain vs. prior, unsurprising people feel things have gotten worse.
@deanwball I think a fair number of elites have updated intellectually, though I’d be curious how many have deeply internalized it. To wit, understanding the underlying drivers of progress and reckoning with the likely consequences. Exponentials are hard!
@ModeledBehavior Generally agree, though at least over the short/medium-term you need to contend with the geographic concentration of datacenters/compute which are both heavily skewed to US & China. Capital, energy, and chip restriction reqs likely to keep this the case for some time.