“Only make chips that can only be used to run inference on existing AI models, or that aren't AI accelerators at all. We call these as inference-only chips.”
LessWrong (Curated & Popular) · "“Hardwired Pause” should likely be your favorite (pause) plan." by David Scott Krueger
Play from 1:05
1:05
Audio streams from the publisher’s own feed. Verbatim does not host or copy it.
What was described
propose using inference-only (hardwired) chips that can only run existing AI models and cannot be used for training
Said nearby on the same subject
“First thing you ask is, are you using my data to train third-party AI models? And it's gotta be a hard no”
AP
Atlas Primer speaker
Using AI at Work: AI in the Workplace & Generative AI for Business Leaders ·
“I used Phoenix, which is a little evals library that you can set up there to run tests, immense real information, and used a large model as a judge to pick the best model that could do the job.”
“The cost to buy NVIDIA's latest GPU chip, Blackwell, can exceed $5,000 per chip. An AI data center can require at least 100,000 Blackwell chips, sometimes more. That means that the cost of just the GPUs is around $500 million.”
SE
Steve Eisman
The Real Eisman Playbook ·
Hard number$5,000 per GPU chip; $500 million for 100,000 GPUsShare this moment
Moments like this, every Monday
The strongest new moments on your topics, in the speakers' own words. Your first brief is free, then USD {{ TIERS.starter.price }} a month for three topics.
No card to start. No password. Cancel any time.
Track a competitor?
Pro watches up to 25 names or phrases and emails you the morning after they are said.