Thanks for the article but I don't think I have the sequence reversed.
Without access to the tracking signal, it's been more important to build out a system that can recreate the value from that lost signal by analyzing what users are actually sharing and saying on their platform. Hence the importance of chat (VR, text, video...) and AI that can be used to process and extract value from a chat signal.
I believe Meta's primary revenue source is still advertising (98%), so that is probably 98% of the why.
> Thanks for the article but I don't think I have the sequence reversed
I suppose I read your first sentence as being in future tense when it might not be. The thrust of my argument is that Meta already successfully built those ad targeting models (Advantage+), and they preceded the Llama releases, so they don't need to use Llama-derived models for ad targeting, as I understood your comment to be suggesting. The sequence was not/will not be "Llama -> ad targeting", but was "ad targeting -> Llama"
Meta didn't have to release the weights of the models. Ad revenue doesn't explain why they did so.
Lots of applications for the technology are likely and already being used.
Llama to improve Customer to Product Matching through content analysis.
Llama for Ad Generation based on customer and product profiles.
Llama to generate additional chat signal for humans not chatting with other humans.
Llama as a site content moderator.
Llama as a code generation assistant.
Llama to clean up and improve internal wikis.
Llama as an operational analysis tool to detect inefficiencies in internal communication...
They will use it every in every situation where they find a successful use case. Open sourcing the weights buys goodwill but it won't be long before everyone open sources their weights since the only real long term moats are the successful products built around the technology and not the technology itself. Facebook is its own moat. Altman suggested this outcome a few days ago in his NYT DealBook interview when he compared the transformer model to the transistor.
Without access to the tracking signal, it's been more important to build out a system that can recreate the value from that lost signal by analyzing what users are actually sharing and saying on their platform. Hence the importance of chat (VR, text, video...) and AI that can be used to process and extract value from a chat signal.
I believe Meta's primary revenue source is still advertising (98%), so that is probably 98% of the why.