I am not sure what open source models are accomplishing another than killing the lead from the competition (openai), only to give it to someone else who has expertise in the area of distribution. This will be yet another good addition to systems like Amazon BedRock.
Many of the recent innovations in both LLM architecture and inference were only made possible through open models such as Llama 2 and Mistral 7B as a starting point for iteration and refinement, which in turn backpropagates (heh) back to the LLMs developers.
It's a win-win for everyone. That's the power of open source.
Well, look at the history. Google had an insurmountable lead, so Elon started OpenAI. Now OpenAI has an insurmountable lead too. So everyone else is starting in third place, or lower. David versus two Goliaths. If you try to become a third Goliath, you'll probably just get smashed. You're later to the game. In this situation, going scorched earth becomes a viable strategy. Slay the Goliaths. Become a hero to the masses. Attract the world's best talent who don't want to be associated with proprietary models. At that point you have a world class AI business with momentum towards AGI. And even if you're giving away last year's technology for free, the team you built is churning out new ideas that could be a financial bonanza one day. Shareholders are willing to pay for a long-term bet if the story is good.
I haven't seen anything about the larger architecture, but I think the value of grok is going to come from it's cheap access to twitter data for RAG etc.
