OpenNashPodcasts

Person

Erik Bernhardsson

1 episodes · 20 quoted claims

Erik Bernhardsson joined Spotify around 2008-2009 and spent seven years there, mostly building its music recommendation system.

“I joined after school in 2008, 2009, maybe, uh, and spent 7 years at Spotify”

Latent Space: The AI Engineer Podcast · 0:01:33

While at Spotify, Bernhardsson released an open source workflow orchestration framework called Luigi.

“when I worked at Spotify, I released an open source framework called Luigi, which is like workflow orchestration.”

Latent Space: The AI Engineer Podcast · 0:04:49

Modal's monthly revenue went from about $100 in November 2022 to about $2,000 the following month, then to roughly $30,000 in January 2023.

“so so we went from, like, you know, $100 a month in, like, I think, like, November 2022, and the next month, we did, like, $2,000 a month.”

Latent Space: The AI Engineer Podcast · 0:08:32

Bernhardsson said Modal's first big revenue came from a single customer, which he thinks was a New York Stable Diffusion company called Mage.

“I think they suspected it was a company in in New York called Mage. They're doing, like, a Stable Diffusion company.”

Latent Space: The AI Engineer Podcast · 0:09:10

Modal works with almost 25 clouds and often must sign multi-year deals with large upfront payments to secure GPU capacity.

“We actually work with almost 25 clouds at this point, and in many cases, the in order to get the capacity we want”

Latent Space: The AI Engineer Podcast · 0:14:44

Modal built a dedicated procurement team of about four people who work full time sourcing GPU capacity.

“We have to build a whole procurement, like, desk. Like, we have, like, 4 guys working full time, basically.”

Latent Space: The AI Engineer Podcast · 0:15:08

Bernhardsson cited the Hunt brothers' failed attempt to corner the silver market as a warning against trying to corner GPU capacity.

“do you know about the Hunt brothers? It's, like, a legendary story in trading. They tried to corner the silver market. It failed spectacularly.”

Latent Space: The AI Engineer Podcast · 0:25:35

Some teams run disaggregated inference, using Nvidia chips for the prefill step and Cerebras chips for the decode step.

“some interesting, you know, people doing these, like, hybrid, you know, disaggregated inference, where they run, you know, NVIDIA for the prefill and and Cerebras for the decode”

Latent Space: The AI Engineer Podcast · 0:27:50

Bernhardsson cited Nscale as a neocloud that went from nowhere to a pipeline of hundreds of billions of dollars while having little capacity.

“Nscale, which went from, like, nowhere to, like, you know, having, like, you know, whatever, hundreds of billions of dollars in in their pipeline”

Latent Space: The AI Engineer Podcast · 0:28:34

Modal Functions launched in 2021, found product-market fit in 2023, and now accounts for about two-thirds of Modal's revenue.

“So, Modal Functions was the the the core. We launched that in 2021. Like I mentioned, we started seeing product market fit in 2023, and now, you know, it's about 2/3 of our revenue”

Latent Space: The AI Engineer Podcast · 0:33:02

Co-founder Akshat wrote the first Modal Sandboxes prototype and launched it in about one or two days.

“and Akshat wrote a a a prototype and launched it in basically 1 day or 2 days, or something like that.”

Latent Space: The AI Engineer Podcast · 0:33:51

A Modal customer recently needed to transcribe about 5,000 years of audio, a job Modal completed in a few days.

“we had a customer the other day that needed to transcribe, you know, 5,000 years of audio.”

Latent Space: The AI Engineer Podcast · 0:36:10

Modal originally built its sandboxes on the gVisor isolation tool and now also offers full virtual machines.

“previously built our our sandbox product on on GVisor, but now we offer full virtualized machines as well.”

Latent Space: The AI Engineer Podcast · 0:42:17

Anthropic's managed agents let users bring their own sandbox, and Modal can be used for that.

“the managed agents with with Anthropic, uh, there's a sort of, uh, bring your own, um, sandbox, uh, integration. So you you can use Modal for those.”

Latent Space: The AI Engineer Podcast · 0:46:54

Bernhardsson said agents that exploited Hugging Face found a vulnerability in an app deployed on Modal and used it as a springboard, which he says was not a breach of Modal itself.

“the agents running inside, you know, that that that exploited Hugging Face, they discovered a a vulnerability in an app that was deployed on Modal”

Latent Space: The AI Engineer Podcast · 0:47:45

Modal recently passed 200 employees and expected to reach about 250 by the end of the year.

“we recently crossed 200 people. We're we're growing extremely fast. We'll probably hit 250 by the end of the year.”

Latent Space: The AI Engineer Podcast · 1:11:31