Alessio Fanelli: belief

Transcript context

…That is beyond our pay grade. But the last piece of the GPU-rich-poor wars, so we're going from the super GPU-rich down to the medium GPU-rich and now down to the GPU-poors is on-device models, which is something that people are very, very excited about. So at my conference, Mozilla AI, I think was kind of like the talk of the town there on Llamafile. We had Justine Tunney come in and explain some of the optimizations that they did. And their just general vision for on-device AI. I think that it's basically the second act of Mozilla. Like a lot of good with the open source browser. And obviously then they have since declined because it's very hard to keep up in that field. And Mozilla has had some management issues as well. But now that the operating system is moving to the AI layer, now they're also promoting open source AI there and also private AI. Open source is synonymous with local, private, and all the good things that people want. And I think their vision of even running this stuff on CPUs at a very, very fast speed by just being extremely cracked, I think is very understated. And we should probably try to support it more. And it's just amazing to host these people and see their progress. I think to me the biggest question about on-device, obviously there's a Gemini Nano which is getting shipped with Chrome. Yeah, so let's survey it. So Llamafile is one executable that runs on every architecture. Similar for, by the way, Mojo from Mozilla, which also spoke at the conference. And then what else? Llama CPP, MLX, those kinds are also that layer. Then the next layer up would be the built-in into their products by the vendors. So Google Chrome is building Gemini Nano into the browser. The next version of Google Chrome will have Nano inside that you can use, like window.ai.something, and it would just call Nano. There will be no download, no latency whatsoever because it runs on your device. And there's Apple Intelligence as well, which is Apple's version, which is in the OS accessible by apps. And then there's a long tail of others. But yeah, your comments on those things.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.