Tech development doesn’t proceed in a straight line for a number of reasons. Currently the leading AI model developers are in a predicament where if they don’t develop as fast as they can, their competitors, including China, may surpass their capabilities and put them out of business.
OTOH, if they do continue to develop as fast as they can, the needed security and alignment features seem to fall behind what the new models can do. AI models getting “out of contain” has many dangers, some of which are basically “game over” for humans.
What to do? Something a bit similar took place with US railroads’ initial buildouts in the 19th Century. Railroad companies had to build new track as fast as they could to beat their competitors: First to get the infrastructure in place usually won the race.
But if everyone was laying track as fast as they could all over the country, inevitably there came a point where there was more track than feasible business models could support. (Something similar has been suggested for what happened during the Fiber Buildout for the late 20th Century internet’s new bandwidth needs.)
Eventually there were RR bankruptcies and mergers and buyouts… and some governmental regulation as well ….until the number of railroads more or less fit what the business model could support. Which also created monopolies and RR barons, and while that’s another topic with echoes in the AI development and DataCenter buildout, that’s a focus for a future post.
Loading...
Reload document 
Open in new tab =========================================================
Full AI Oversight Report 9/12/26
Loading...