Two sentence horror:
For rent: GB300s, never used.
For rent: GB300s, never used.
Heard from a neolab friend:
“Even if we fail to train the best model, we locked in GPU capacity early. H100/B200/B300 rental prices have since gone up 2–3x. We can just resell the compute and print money.”
OK, so that’s the business model.
19
22
958
67,612
It’s amazing you can clearly tell exact moment in history when each lab started vibecodong their product
19
3
556
30,366
Playlist this week:
- upping my p(doom)
- chasing the roofline
- core auto thesis (internal)
- upping my p(doom)
- chasing the roofline
- core auto thesis (internal)
8
0
114
9,242
It is our responsibility to make reality interesting, as otherwise humans will rationally choose to live in a simulation
20
12
207
11,638
Oh, hi!
We think computational depth is the missing scaling axis, i.e. we should be doing a lot more deep learning!
Every other axis has been scaled by OOMs over the past few years (params, data, sparsity, test-time reasoning), but depth has been stuck at ~100 layers since GPT-3.
We've h *severely* depth-bottlenecked and bad at using the depth they have, and that architectural interventions that lift this bottleneck efficiently lead to gains that increase with compute.
w/ @akshayvegesna
6
8
239
29,968