The Information Bottleneck
The Information Bottleneck
What a VC Actually Looks For in AI Founders | Sarah Catanzaro
0:00
-1:05:13

What a VC Actually Looks For in AI Founders | Sarah Catanzaro

In this episode, we talked with Sarah Catanzaro, a general partner at Amplify Partners, about how AI research looks from the investor's side. Sarah has backed Runway and Modal since their seed rounds, and she usually gets to know researchers long before they start a company. This one is a bit different from our usual episodes because we talk about business more than research.

We started with what makes a research idea a good business. Sarah's view is that startups shouldn't compete with the frontier labs directly and should go after things the labs put lower on their list. Voice is her example. Early experiments suggested that adding audio data hurt reasoning, so the labs didn't prioritize it and companies like ElevenLabs got ahead. She thinks the new auto research startups are making a similar bet.

We also talked about AI slop. Sarah invested in Taste Labs, which is trying to end slop first and leave the question of good taste for later, because people agree much more easily on what is bad. From there we got into benchmarks. She likes public benchmarks, but when we asked if a benchmark win impresses her in a pitch, she said "I don't care." What she looks at is whether people adopt the product and whether deals close.

We spent a lot of time on founders. Sarah wants a founder to have a big vision and a concrete first problem to solve, and she stays away from what she calls "research" for a problem. She told us you can't evaluate someone from a resume, so she reads their papers and talks to them.

On funding, Sarah thinks the growth in AI is real, but early-stage neolabs look bubbly to her. By her count, close to a dozen auto research companies have raised more than $100 million each, and none of them has revenue. She explains why founders have about two years to show progress and why she tells her companies to pay top dollar for people. She is also critical of tranche rounds announced at the second-tranche valuation. Near the end, we talk about spending on tokens vs people, whether fame helps a founder, investor FOMO, and the SF echo chamber.

Topics

  • When great research is also a great business

  • Where startups can beat the frontier labs

  • AI slop and benchmarking taste

  • Public benchmarks vs closing deals

  • What Sarah looks for in founders

  • Is AI a bubble? Neolabs and auto research

  • How much to raise and how to spend it

  • Does fame help a founder, and investor FOMO

Timestamp

0:00 Intro
0:44 Great research vs a great business
6:04 Where startups can beat the frontier labs
8:54 Regulation and "capabilities maxing"
12:10 AI slop and benchmarking taste
21:44 Do benchmarks impress investors?
25:46 How to get an investor's attention
30:29 What Sarah looks for in founders
37:23 Is AI a bubble?
40:55 How much to raise and how to spend it
46:29 Tokens vs people
50:57 Does fame matter?
57:08 FOMO, the SF echo chamber, enterprise vs consumer
1:03:32 Best advice for founders

Music

* "Kid Kodi" - Blue Dot Sessions - via Free Music Archive - CC BY-NC 4.0

Discussion about this episode

User's avatar

Ready for more?