

Trying to etch models into a chip is a dead end until we reach “peak” quality.
However, unless they include some kind of LoRA (low-rank adaptation) adapter onto the silicon, it severely limits the utility of whatever model or architecture they choose. Being able to modify the weights is way more useful.
Honestly, diffusion decoders are probably where we’ll end up some day. Not end there, but that’s probably the next logical step in the throughput chain.
General purpose compute is infinitely more valuable during times of great software improvements than highly specialized compute.
Things like Tensor Processing Units (TPUs) still aren’t ubiquitous yet, even though they’ve been around for 10+ years. They’re Too specialized to allow for reasonable flexibility on testing.





That 8 hour 20 minute spec for returning to baseline could probably be sped up a bit to improve the inter-frame timing
Modern monitors measure gray-to-gray time as the response, rather than black-to-black. So if she computed the 50% luminescence threshold time on the decay side, and then started injecting the next round at that time, could likely cut the decay time to closer to 200 minutes minutes for the first frame, and then probably double that for the inter-frame times, depending on the GFP decay rate.
Who knows, maybe she already took that into account. She seems to be well rounded.
From the charts in the video (above), it looks to be symmetrical near the peak, but the 50% concentration on decay is around 200 minutes minutes of decay followed by 70 minutes ramping up would put the inter-frame time to be about 3.5 hours.
I realize concentration of fluorescing enzyme may not be 1:1 with lumens emitted.
If this worked the way I imagine, This cuts the total time to play from 600 years to about 200 years. Rounded to 1 significant figure, since this is all ballpark math anyway.