“To use the GPU power more efficiently, you can aggregate the API code from different users to fill in the computes.”
ML
Ming-Yu Liu Nvidia
Practical AI ·
“…hard to justify me taking up hundreds of GPUs for a substantial amount of time … unless they’re in line with the company's goals or there’s downtime and underutilized GPUs.”
Eric Bigelow, Goodfire · "The Cognitive Revolution" · How Agents Decide: Goodfire's Eric Bigelow on Critical Tokens, Phase Shifts, & In-Context Learning
Play from 47:55
47:55
Audio streams from the publisher’s own feed. Verbatim does not host or copy it.
What was described
Allocate GPU time to experiments only when they align with company goals or when there is idle compute, to avoid using hundreds of GPUs for long runs that aren't justified
Said nearby on the same subject
“If you're a neocloud, you can buy the GPUs, deploy them, sell a strip of GPU futures, and then just sell on-demand compute…sell six month contracts…post a big percentage upfront”
♫
Said on Odd Lots
“I made an RL agent that really utilized GPUs. I could scale like linear with compute… it was very efficient and easy to extend.”
♫
Said on Training Data
“Even at three agents, what we found was three agents were not aligning on a simple goal more than 16% of the time.”
VP
Vijoy Pandey Cisco
DataFramed ·
Moments like this, every Monday
The strongest new moments on your topics, in the speakers' own words. Free for three topics.
No card. No password. One-click unsubscribe.
Track a competitor?
Pro watches up to 25 names or phrases and emails you the morning after they are said.
See Pro, $29 a monthVerbatim quotes a short extract and links to the episode; the episode belongs to its publisher. Remove a show.