Why open source AI models?

[TIJ]

Why are big tech companies open sourcing AI models. What do they hope to gain? Unlike in case of Linux and Linux foundation, these corporations are for profit and it takes an unusually large sums of money to train these models?

[rvz]

The following reasons:

1. To not pay for fully closed-source models and have it close to free.

2. It makes it possible for intelligence at the cost of less than $500, significantly less than $50k for a writer and much less than $250k for a SWE.

3. Have it local and private.

The race to zero for intelligence will be accelerated and will affect everyone. If we all know that "AGI" is a scam, then at least it should be given to not one single company and should not be re-defined to mean something else.

[blackeyeblitzar]

It’s just Meta and their goal is to not look like they’re behind and to deny other tech companies a profit margin by just offering it for free, since they don’t have a real way to monetize it directly.

Also these aren’t open source, just open weights. AI2 is the only real open source model that’s well known

https://allenai.org/olmo

[talldayo]

> It’s just Meta

And Google, then after a while Microsoft and Apple too.

Meta fired the starting gun, but everyone else sheepishly had to admit that they had no moat either and coughed up their models for free to appear at "parity" with a company as reprehensible as Meta. After all, you can't pretend to be morally superior in a market where your competitor has the curtain pulled all the way back. OpenAI's stock practically correlated to Meta's press releases after a certain point.

You have to admit Meta played their cards right. Microsoft and Apple would have ruined the product design of AI a-la Cortana and Siri, and Google was giving their AI away since the Deepmind days. If any of their competitors were ahead of them then they entirely failed to produce a competitive example. Or a superior business model, for that matter.