The AI Race
In a previous post I asserted that Google is losing the AI race because of Innovators Dilemma and Engineering-Centric Search prioritized over Human-Centric Chat. I think Meta has a leg up over all AI companies, including OpenAI, Anthropic, Hugging Faces, etc., because:
1) Facebook/Instagram has 3B+ users giving them social data for over 15 years
2) Meta open-sourced (open-weighted?) Llama3
3) Social Media is inherently more human-centric than Search. But then Facebook acquired WhatsApp 10 years ago. This seemed like a foolish $19B move then, but now it's brilliant: Chat is more human-centric than social media which is more human-centric than search. My old Yahoo! boss Brian Acton empowered Zuckerberg/Meta with a billion chat users and their data, which has grown to 2B WhatsApp users now.
On the topic of chat, China has been chat-centric for even longer than the US. Tencent QQ has been around since 1999, and unlike ICQ and AIM, QQ had human-centric features incorporated into chat from the beginning, like commerce transactions, stickers, virtual goods, gaming, booking hotels/flights, etc. Chinese have been living their everyday life on Chat for a long time. Chinese also have more human data than us: 5x US size and a population that is conditioned to not care about privacy and surveillance. China also has the brain trust; all their PhDs that they got degrees here and worked in AI Tech companies here, are or have moved back - the repatriation brain drain (e.g. ex-Yahoo! Qi Lu). If China has access to the latest GPU from Nvidia, China would win the AI race - this is why the US government has been regulating Nvidia GPU sales to China. The AI Race is the new Space Race or Arms Race, and it has massive implications on the world’s two remaining SuperPowers.
i am not sure about China winning the LLM race. For one, when Chinese government said they supported AI, they meant industrial AI, not chat or entertainment focused AI like GenAI. i know for a fact Huawei and other chinese companies are looking to sponsor AI researches in US focusing away from a while back. (obviously that is no longer possible now). The argument for industrial AI is that China has FAR more industrial data than US given the size of Chinese manufacturing industry AND industrial AI would solve the upcoming demographic problem.
i have been following AI driving for a while, and i have experienced chinese AI driving in real life. i have to see their performance appears to be significantly better than tesla. One reason is the use of multiple sensors in AI including LIDAR which china has managed to bring the cost down so much it costs something like 700 bucks to have LIDAR all over the car. in the meantime, all US LIDAR companies are basically dead or on life support. i was very surprised US gov didn’t offer any help given the obvious military use of LIDAR.
I think one other reason why Meta is making Llama3 open source (more like open weighted) is about monetization. As the three large cloud infrastructure providers Microsoft is backing OpenAI, Amazon is backing Anthropic and Google has Gemini and they all come as pay per token service. This is a favorable set up for cloud providers. Meta doesn't have infrastructure monetizing business model. So, for Meta open sourcing their foundation model is a preemptive strike against the others to weaken their hand.