Today’s episode of Ai Unchained is on this and while a lot of what they showed is incredibly impressive, I take the show to separate what I think are the real breakthroughs and what is the bullshit.
Also the major elements here I think are implementation and multimodal innovations, not actually vast improvements in the models themselves. nostr:note1tfjehz6np6awdldrjz9yfrsrvuhphg6h2z9kn2fmlmu2ggegcs3qqwvm5x
Apple did that with...everything. Other companies focused on hardware specs, Apple optimized it for grandma.
I didn’t see open AI’s but Google was sick. Problem is just that I don’t trust Google. Can’t use it. Too much data for them
Faster, cheaper, *marginally* better, and its speech to speech unified not a carousel of different models like before (whisper->gpt->tts)
Haters acting like it isn't a huge breakthrough
The UX of this is incredible and is basically revealing that the promise of all of this that has been since we first got a glimpse of ChatGPT almost 2 years ago now is completely possible.
However, the fact that the major breakthroughs are in how all of these tools have been put together, rather than in some new 3Trillion param model or some shit is actually really reassuring. The models we have now, if used and leveraged correctly, can already do 100x what we think they can in their “naked” mode (the base LLM chat)
