To argue the other side:

1. It's fundamentally atoms-limited by GPU capacity?

2. Attention is still scarce / fickle?

3. Regulatory capture / interfacing with analog / bureaucratic systems still exist?

1. disagree. imo w/ recent SoC and tpu endeavours, these are less physical and more manufacturing constraints. I might be wrong b/c limited chip design knowledge

2. agree. hence my extended comment which Chamath (and most) have not realized: new lifespan of apps= ~ephermal ie bullish FC protocol!

3. 💯^2

I would say a lot of people are gonna learn a lot about the Gartner Hype Cycle. It's 100% true LLMs will have massive impacts but their impacts are being overstated right now. Current limits are context sizes and those will be solved. Next steps will be wrangling LLMs through meaning prompt crafting for targeted output