Feels like 50/50 Anthropic nor OpenAI exist as independent companies in five years. I remember having this debate that someone was incensed I thought AI might not produce a FAANG company, but people don’t tend to buy pure technology.
In a weird way, looking at the existing companies, it seems “obvious” their struggle will be real. Google is an advertising company, so is Facebook. Most of apples competitors, and Apple itself sell other products. And it’s funny they’re building phones when I think Apple itself is starting to realize the iPhone’s time as a gazillion dollar maker is coming to an end.
100B isn't too far fetched. I can run 27B-iq3s on my Pixel 10 Pro XL (it'll run larger models but at slower speed). It depends on the quant and the number of active parameters, since MOE models are the obvious choice for a mobile device. Something like 100B-A10B would be really nice if it can run reasonably fast. Mobile GPUs will get a boost along the way no doubt.
Feels like 50/50 Anthropic nor OpenAI exist as independent companies in five years. I remember having this debate that someone was incensed I thought AI might not produce a FAANG company, but people don’t tend to buy pure technology.
In a weird way, looking at the existing companies, it seems “obvious” their struggle will be real. Google is an advertising company, so is Facebook. Most of apples competitors, and Apple itself sell other products. And it’s funny they’re building phones when I think Apple itself is starting to realize the iPhone’s time as a gazillion dollar maker is coming to an end.
Turns out making trillions of dollars is hard.
100B isn't too far fetched. I can run 27B-iq3s on my Pixel 10 Pro XL (it'll run larger models but at slower speed). It depends on the quant and the number of active parameters, since MOE models are the obvious choice for a mobile device. Something like 100B-A10B would be really nice if it can run reasonably fast. Mobile GPUs will get a boost along the way no doubt.