Adding AI researchers decreases group productivity when GPU capacity is constrained
Logan Kilpatrick · Inside OpenAI | Logan Kilpatrick (head of developer relations) · Feb 8, 2024 · at 43:45
Logan Kilpatrick, Head of Developer Relations at OpenAI, explains why scaling headcount on AI research teams differs from scaling product engineering teams.
0:00 / 0:34exact quote · 34.1s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“In a world where you're constrained by the amount of GPU capacity that you have as a researcher, which is the case for OpenAI researchers, but also researchers everywhere else, like Each new researcher that you add is actually, like, a net productivity loss for the research group, unless that person is, like, up-leveling everyone else in, like, such a profound way that, like, it increases the efficiency. Like, if you just add somebody who's gonna go and, like, tackle some completely different research direction, you now have to share your GPUs with that person, and everyone else is now slower on their experiments.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Logan Kilpatrick
OpenAI models will likely never match vertical tools like Harvey
“Our models are probably never going to be as capable as some of the things that Harvey's doing because like our goal and our mission is really to solve this like very general use case.”
Startups building general-purpose AI agents will directly compete with OpenAI
“I talked to a lot of developers who are building like, you know, just general purpose assistance and like general purpose agents and stuff like that. And I think it's cool and it's a good idea. I think like the challenge for them is like, they are going to end…”
GPT-5 will normalize as a useful tool rather than changing everything
“GPT-V is like surely going to be extremely useful and like solve some whole new echelon of Problems. Hopefully they'll be faster. Hopefully it'll be better on all these ways, but like fundamentally the same problems that exist in the world are still going to b…”
Kilpatrick: A board crisis in 5-10 years would be far more damaging
“Fast forward like five or 10 years of something like this would have happened. And we sort of hadn't gone through the hopeful upcoming like word transformation and sort of all those changes that are going to happen. I think it would have been a little bit or p…”
OpenAI will not build highly verticalized AI products like sales agents
“Versus like, we're not going to launch like some of these like very verticalized products. Like we're not going to launch like an AI sales agent. Like that's just not what we're building towards.”
OpenAI views revenue merely as a proxy to acquire more compute
“Even if revenue is a goal, it's like revenue is not actually the goal. Revenue is a proxy for getting more compute, which is then like actually what helps us get towards getting more GPUs so that we can, you know, train better models and like actually get to t…”