The path to ubiquitous AI (17k tokens/sec) https://news.ycombinator.com/item?id=47086181
replyYou can still try it at https://chatjimmy.ai/, but it's running the rather outdated Llama 3.1 8B
Yes, from 6 days ago: https://news.ycombinator.com/item?id=48986351
replyThat looks promising! As models become a commodity, this may turn out to be the real AI gold rush.
replyA question of course would be "is 15,000 tok/s Gemini better than 100 tok/s Opus 5"?
replyIt's better for the billions of free users that Google serves.
replyDepends what you do. We have certain tasks we spend money on where Gemini 4.6 definitely is better than Opus 5.
replyGiven how fast models are improving, burning the weights into actual ROM is prohibitively expensive if you need (or want) to replace the chip every couple of months.
replyThe alternative is on-TPU flash for storing the weights.
Yes. We're still a ways off from this being ubiquitous, but I am convinced it's coming.
reply