if straight lines on graphs are to be believed, this is the last year in which agi labs are merely highly resourced megacorporations. this is the year of cyberpunk. 2027 is probably too weird for even that.
i have mostly found the opposite of this to be the case so far? ai has replaced much of my demand for specialists, but still needs to be driven by a capable generalist. domain expertise is the thing it's best at!
It’s underappreciated how radically different Jensen Huang’s view is of how AI progress is likely to go than the view of other AI CEOs.
For instance, he recently claimed that AGI will never replace domain expertise.
aiui, growth historically has come from 3 things:
1. capital investment which makes the existing quantity of labor more productive
2. better coordination allocating existing labor more efficiently
3. population growth increasing the total amount of labor
effects on 1 & 2 are arguable but if you replace "AGI" with "uploaded human minds placed in humanoid robot bodies" you could double the population of said robots roughly every year. it's just much faster to build computers & robots than to grow humans.
Can diffusion models generate high-quality images without CFG?
Yes! We introduce ✨Correction Training for Diffusion Models✨, which significantly improves sample quality without relying on classifier-free guidance.
Project page: https://t.co/Ic8foxITJ0
🧵 (1/n)
@DavidSHolz custom, claude wrote it when @_sholtodouglas asked it to, iirc he basically just asked for a cool visualization of the giant evolving codebase
Green button = You instantly receive $2 million (tax free).
Red button = You go back in time and restart your life at 10 years of age, with all the knowledge you have today.
You can press just once. Which do you press?
🧵Scaling Laws for T2I Diffusion
We do a scaling laws ladder for t2i diffusion models all the way from 60M to 2B, spanning 3 orders of magnitude in training flops
1y from today this will no longer be true. even today i'd say it's more of a skill issue than a model issue but this time next year it'll just be overwhelmingly false.