STShubhendu Trivedi@_onionesque4:51 PM · Jul 19, 2026

Now you have a lot of usually laughable philosophy around closed (in terms of safety, risk, security, sovereignty). That too has many reasons, but a lot can be traced back to folk who were essentially researchers who have the delusion that capability is upstream of monetization.

STShubhendu Trivedi@_onionesque4:32 PM · Jul 19, 2026

1718TECH

One of the major reasons that ML (now it's all AI) really took off around '11-13 was because of its unusually open culture [of "frictionless reproducibility" to borrow from Donoho]. There was a Big Grab around '18-21, and thus was birthed the aberrant discussion around closed AI.

STShubhendu Trivedi@_onionesque4:43 PM · Jul 19, 2026

1718TECH

There were of course many other reasons and coincidences that were direct causal factors for the acceleration. e.g. after Dan Ciresan's use of GPUs for v.simple feed forward models, people saw the potential to scale, and just then a lot of post GFC stimulus money was available.

STShubhendu Trivedi@_onionesque4:46 PM · Jul 19, 2026

1718TECH

But the key factor was that it was always open (even Transformers, for example). Enough capital and enough ambient know-how had been accumulated to start thinking of them as reproducible capital assets.