Raw LLM Responses

I think ByteDance is able to produce such a good video model simply because they have such a massive amount of video data from TikTok to train on, just like Google DeepMind has with YouTube. I don’t think this translates to being like Google DeepMind in that they are closer to AGI than most competitors, because this is determined by the amount of compute, high-quality data, and top AI research talent density a company has at their disposal. Even the data part is dubious since synthetic data is being more heavily relied on and actually produces very capable models, so it’s really just compute and talent. This lines up with Google DeepMind and OpenAI capital expenditure, which seems to be going mostly to datacenters and recruiting top AI researchers. ByteDance is probably trying to build datacenters but will struggle for obvious reasons being a Chinese company. And for the same reasons that many of the top AI researchers choose to not work at Meta and xAI, most of these top talents probably won’t want to spend their limited pre-AGI time working on improving TikTok’s addictiveness or anything else they view as a net negative for humanity. Not all of them, mind you, but quite a sizable amount of them