3 ms·

> But I suspect it's also a case of "If we can't dominate AI, no one must dominate AI". Embrace. Extend. Extinguish.

lcnPylGDnU4H9OF 2y ago
What they're doing is almost literally the opposite of EEE. If OpenAI actually had open models, then Facebook could take those models, add their own non-open things to the models, and use this new stuff as a business advantage over OpenAI. Instead, they're independently developing their own models and releasing them as open source to lower the value of proprietary models. https://en.wikipedia.org/wiki/Embrace,_extend,_and_extinguish https://en.wikipedia.org/wiki/Embrace,_extend,_and_extinguis...
- yazzku 2y ago
  Good point against the EEE case above, but the models are not open source per the OSI definition that everyone understands. That part is part of their marketing playbook. Licenses are almost always non-commercial use, which makes them not-open. You could say it's "open model", but that's a useless term since the arch is not a secret, but rather the training data that went into it, which is not open.
- mywittyname 2y ago
  What you're calling EEE is different from how that wikipedia page defines it... > Phase 1 (Embrace): all participants need to establish a solid understanding of the infostructure and the community—determine the needs and the trends of the user base. Facebook built their own independent models and platforms. Check. > Phase 2 (Extend): Offer well-integrated tools and services compatible with established and popular standards that have been developed in the [...] community Check. > Phase 3 (Innovate): move into a leadership role with new [...] standards as appropriate[...] Change the rules: [LLaMa] become the next-generation [LLM] tool of the future. Sounds like that might be their goal. I'm curious which phase you think Facebook avoided and why? Internet Explorer "lowered the value of propriety" web browsers by being included with Windows at not additional cost. So lowering the value of a competitor is definitely in the EEE playbook. I'd go so far as to argue it's a pillar of what makes the playbook so effective.

What they're doing is almost literally the opposite of EEE. If OpenAI actually had open models, then Facebook could take those models, add their own non-open things to the models, and use this new stuff as a business advantage over OpenAI. Instead, they're independently developing their own models and releasing them as open source to lower the value of proprietary models. https://en.wikipedia.org/wiki/Embrace,_extend,_and_extinguish https://en.wikipedia.org/wiki/Embrace,_extend,_and_extinguis...

yazzku 2y ago
Good point against the EEE case above, but the models are not open source per the OSI definition that everyone understands. That part is part of their marketing playbook. Licenses are almost always non-commercial use, which makes them not-open. You could say it's "open model", but that's a useless term since the arch is not a secret, but rather the training data that went into it, which is not open.
mywittyname 2y ago
What you're calling EEE is different from how that wikipedia page defines it... > Phase 1 (Embrace): all participants need to establish a solid understanding of the infostructure and the community—determine the needs and the trends of the user base. Facebook built their own independent models and platforms. Check. > Phase 2 (Extend): Offer well-integrated tools and services compatible with established and popular standards that have been developed in the [...] community Check. > Phase 3 (Innovate): move into a leadership role with new [...] standards as appropriate[...] Change the rules: [LLaMa] become the next-generation [LLM] tool of the future. Sounds like that might be their goal. I'm curious which phase you think Facebook avoided and why? Internet Explorer "lowered the value of propriety" web browsers by being included with Windows at not additional cost. So lowering the value of a competitor is definitely in the EEE playbook. I'd go so far as to argue it's a pillar of what makes the playbook so effective.