scripod.com

Jonathan Ross, Founder of Groq

David Senra

6 DAYS AGO
David Senra

David Senra

6 DAYS AGO
Jonathan Ross, founder of Groq and inventor of the Google TPU, discusses his journey from creating a custom AI chip at Google to leading a major partnership with NVIDIA. He shares insights on leadership, the importance of fast inference for AI, and the challenges of building a company against conventional wisdom.
Ross explains that combining GPUs and LPUs optimizes LLM processing, leading to a $20 billion partnership with NVIDIA. He emphasizes that faster inference makes models smarter, as demonstrated by AlphaGo's improvement with TPUs. Ross values 'Reality Quotient' over IQ, focusing on identifying the dominant game to play. He discusses 'manufactured discontent' as a driver for innovation, feeling a personal responsibility to provide enough compute to accelerate cures for diseases. Ross shares leadership lessons, including using 'I intend to' statements to invite feedback without pessimism, and the importance of under-constrained goals to foster innovation. He recounts how Groq bonds saved the company by having employees exchange salary for equity, and how he shifted from seeking positive traits in hiring to avoiding negative ones. Ross also discusses the herd mentality of West Coast VCs and how East Coast funds ultimately invested in Groq.
00:25
00:25
Combining GPUs and LPUs defeats various bottlenecks.
01:46
01:46
Current payment systems are not built for the micropayments that will skyrocket
03:30
03:30
A personalized daily brief summarizes news based on his interests.
05:55
05:55
AI shifts success from answering to asking the right questions
08:23
08:23
Leadership is having followers.
13:30
13:30
Set a simple, clear goal for everyone
14:34
14:34
Minimize constraints to foster innovation
19:06
19:06
Good founders ask insightful questions that make experts reconsider their assumptions.
19:44
19:44
His decisions were correct but lacked conviction.
22:23
22:23
West Coast VCs follow each other like lemmings
23:50
23:50
Extra money is no longer an advantage for startups.
26:48
26:48
Combining LPUs and GPUs optimizes LLM inference
30:08
30:08
Faster inference enables deeper search and creativity.
34:57
34:57
Reality Quotient involves recognizing reality and choosing the dominant game to play.
35:55
35:55
Align the entire organization around one dominant metric.
37:11
37:11
Make changes feel like no change at all.
41:06
41:06
Doing things differently is the only way to have an advantage.
42:54
42:54
People only understood the value of fast inference after trying it themselves.
46:32
46:32
Intent invites feedback without pessimism
51:07
51:07
80% participated, many taking large pay cuts
54:19
54:19
Focus on positive traits to avoid hiring problems.
58:46
58:46
Loss feels more painful than gain
1:00:37
1:00:37
Humiliation as a motivator for superhuman performance
1:03:13
1:03:13
Manufactured discontent is necessary for innovation
1:05:02
1:05:02
AI can cure diseases and slow aging.
1:07:11
1:07:11
Code creation is becoming nearly free.
1:10:04
1:10:04
Education should focus on asking questions, not answering them.