Nowadays the "crazy uncle" is the one that denies that his TV spies on him or what ever. Sitting there at the family dinner, telling you to trust Google products. Etc.

I would like trained LLMs to be regarded as some sort of pre-compiled binary. I.e. to have the LLM, without being theoretically able to reproduce it, is not regarded as "open source" at all.

I wonder if it is possible to test if a given data input would end up with kinda the precompiled model, without wasting millions of dollars.

"OpenAI" would imply that it is open if the plebs are allowed to use it.

> I would like trained LLMs to be regarded as some sort of pre-compiled binary. I.e. to have the LLM, without being theoretically able to reproduce it, is not regarded as "open source" at all.

I like this mental model. Don't mind if I reuse it :)

The latter want to find a "consensus", not to defend the user:

"By involving a broad spectrum of stakeholders, we’ve ensured that the Open Source AI Definition reflects a multitude of needs and concerns."

https://news.ycombinator.com/item?id=41968322