Erik Bernhardsson (@bernhardsson) 's Twitter Profile
Erik Bernhardsson

@bernhardsson

Building @modal_labs when I'm not posting bangers about data and software. Previously built the music rec sys at Spotify and ran the eng team at Better.

ID: 1191207924

linkhttps://erikbern.com/ calendar_today17-02-2013 21:32:04

12,12K Tweet

39,39K Followers

3,3K Following

Erik Bernhardsson (@bernhardsson) 's Twitter Profile Photo

There is a rumor Modal is raising. We are! – We are raising the bar for developer experience. Not raising a round though. Don't ping me about it lol.

Modal (@modal_labs) 's Twitter Profile Photo

Happy #NationalInternDay! 💚 Our summer interns at Modal are tackling the hardest challenges in distributed compute – kernel-level packet filters, RAM checkpointing and low-latency inference! 🤓

Happy #NationalInternDay! 💚

Our summer interns at Modal are tackling the hardest challenges in distributed compute – kernel-level packet filters, RAM checkpointing and low-latency inference! 🤓
Charles 🎉 Frye (@charles_irl) 's Twitter Profile Photo

at tomorrow's inaugural Modal webinar, I will show you how I'm running an OpenAI-compatible LLM service that can handle a burst from 100 to 1000 simultaneous users using LLaMA 3.1 8B in vLLM on Modal -- and give you the code to do it yourself!

at tomorrow's inaugural <a href="/modal_labs/">Modal</a> webinar, I will show you how I'm running an OpenAI-compatible LLM service that can handle a burst from 100 to 1000 simultaneous users using LLaMA 3.1 8B in <a href="/vllm_project/">vLLM</a> on Modal -- and give you the code to do it yourself!
Erik Bernhardsson (@bernhardsson) 's Twitter Profile Photo

Yes. I actually think it’s a feature of interviewing (at least when it’s an interviewee’s market) that it’s labor intensive to interview. It signals that the employer is committed and there’s a decent likelihood of an offer.

Erik Bernhardsson (@bernhardsson) 's Twitter Profile Photo

Modal blog post by Charles 🎉 Frye replicating the “Large Language Monkeys” paper: it turns out a simpler model with a more complex search during inference can outperform a more simplex model modal.com/blog/llama-hum…

Erik Bernhardsson (@bernhardsson) 's Twitter Profile Photo

Hearing this from a competitor I used to fear and respect a ton made me slightly tear eyed. And fun fact is that GPU was totally an afterthought!

Erik Bernhardsson (@bernhardsson) 's Twitter Profile Photo

Starting to think that Twitter last few months is falling for the social media platform trap of optimizing for engagement (which is easy to measure and has fast feedback loops) at the cost of long term retention (which takes much longer to measure)

Erik Bernhardsson (@bernhardsson) 's Twitter Profile Photo

Every SaaS company is just a thin wrapper over cloud providers. Every cloud provider is just a thin wrapper over hardware providers. Hardware providers are just a thin wrapper over fabs. Fabs are just a wrapper over silicon miners.

Every SaaS company is just a thin wrapper over cloud providers. Every cloud provider is just a thin wrapper over hardware providers. Hardware providers are just a thin wrapper over fabs. Fabs are just a wrapper over silicon miners.
Erik Bernhardsson (@bernhardsson) 's Twitter Profile Photo

I half-seriously think b2b startups shouldn't be allowed to sell contracts longer than a month for the first few years. The constant threat of churn forces you to build a better product.

Lina Colucci (@lina_colucci) 's Twitter Profile Photo

Can you guess when we posted on Hacker News?? 1000+ videos per hour so far 🤯 Thanks Modal for making scaling a breeze! All we need to worry about is responding to the HN comments, not keeping our GPUs running

Can you guess when we posted on <a href="/hackernews/">Hacker News</a>?? 

1000+ videos per hour so far 🤯

Thanks <a href="/modal_labs/">Modal</a> for making scaling a breeze! All we need to worry about is responding to the HN comments, not keeping our GPUs running