Post #4497844

2026-08-11 11:25 UTC

Lets assume most of the uses for LLMs that are popular now continue to be used with linear growth. So, there isn't any mass public rejection of the tech altogether, but the exponential explosion never happens. Do we need to build data centers to make this possible? The current built-out is for projected explosive growth. Which is silly and financially wishful reasoning. But what would sane adoption look like? 1/

Replies (2)

@futurebird@sauropods.win 2026-08-11 11:29

I guess in a way this is the central question, but the hype is so extreme that I think it's not discuss directly enough? It's hard to abide all of the noise that is both trying to create explosive demand for "compute" ... while at the same time crying that there isn't enough. Seems like many of the big firms have more than they need as they are renting out the excess. 2/

Open ##4497843
@petealexharris@mastodon.scot 2026-08-11 11:32

@futurebird@sauropods.win To me, sane adoption would look like resilient disaggregation into local small language models tailored for most use cases, with general-purpose computing data centres like AWS and Azure (but ideally independent or nationalised ones too), rather than super-resource-intensive single-purpose data centres for large unreliable general-purpose "AI".

Open ##4503123