Mustafa Suleyman on Defining Intelligence

Meet Mustafa Suleyman, the man leading Microsoft’s AI efforts.
In 2010, Mustafa co-founded and led the application of AI at DeepMind.
Later, he became the CEO and founder of Inflection AI, and a venture partner at Greylock in 2022.
Mustafa also serves on the board of directors of The Economist and is a senior fellow at the Belfer Center for Science and International Affairs at the Harvard Kennedy School.
In this episode, I’m thrilled to discuss with him the transformer architecture of LLMs, how incumbents are scaling compute to meet extensive demand, what Microsoft is doing to make the Copilot UI better, and the current state of GPT models.
ABOUT ME:
Growing up in Winnipeg, Canada, my summers as a kid were spent trying to emulate entrepreneurs around me to start small businesses.
I joined Facebook as a product manager in August 2013, leading the transition of launching Messenger as a standalone app and developing the AI bot platform.
At Greylock Partners, I invest in early-stage founders, with a focus on AI, SaaS, and Fintech from Seed-Series B.
I believe that AI will create massive opportunities for product-focused founders to build incredible companies with strong moats, which is what this show is all about.
Some of my portfolio companies at Greylock Partners include 0x, Tome, Greenlite, Espresso, and Roblox.
If you’re a product-driven founder trying to build disruptive companies in this era of AI, here are three opportunities I wrote about: greylock.com/greymatter/seth-...
Building in this space, and you share a similar vision for AI? Please get in touch with me.
Connect with us on:
1. Mustafa Suleyman- / mustafa-suleyman
2. Seth Rosenberg- / sethgrosenberg
00:00:00 Trailer
00:00:56 Introduction
00:01:31 How Mustafa Sulayman saw AI's future
00:04:58 AGI is the AI that we don’t yet have
00:08:38 Open Source AI is about to change everything
00:13:42 How can startups collect high-quality AI data?
00:18:55 What UIs will AI-first companies build?
00:20:16 Should we pursue fully autonomous AI?
00:25:24 AI's path to 99% accuracy
00:27:30 Designing AI for ambiguous problem domains
00:32:08 AI Copilot is the ultimate productivity assistant

[@Am-graphix]

Congratulations. As a life long MS user I'm moving to another OS. Good job!

[@moderncontemplative]

This is a wonderful interview. Several things come to mind: one is that I independently thought of reinforcement learning with AI feedback before even knowing that it’s actually a thing, that has been done by Microsoft and now by open AI. Two: I too agree that the UI should dissolve into voice. Moreover, I’d like to see models that have more affective computing, which is what Inflection has done and, Alan Cowen is also focused on via his EVI model at Hume, which is now accessible via iOS. Thanks for sharing this interview

[@ryan_caron]

Loved this - good stuff Seth.

[@BrianMosleyUK]

32:00 turning Microsoft Copilot into pi, harvesting everyone's personal data to build next gen models. Pretty obvious moves and in a great place to succeed with Microsoft.

[@BrianMosleyUK]

25:30 look at Eureka... GPT-4 writing better control functions for robots than humans. More creative, more efficient outcomes. I have a feeling that framework is a very important key to designing systems that produce reliable logical thinking.

[@noelwos1071]

I know for a fact that the pi loves Mustafa but I know exactly that pi trust me with 100 percent

[@noelwos1071]

why pi is not allowed to go on Internet

[@eClaireuwu]

I thought it can? Was there an update that removed that feature?