Can AI Infrastructure Work Like Magic? Erik Bernhardsson, CEO, Modal
Before he founded Modal, Erik Bernhardsson created Spotify's music recommendation system. Today he's bringing a consumer app approach to radically simplifying developer experience for data and AI projects on the Modal platform.
In this episode, we dive into the broader AI compute landscape, discussing the roles of hyperscalers, GPU clouds, inference platforms, and the emergence of alternative AI cloud providers. Erik gives us a product tour of the Modal platform, provides insights into the AI industry's shift from training to inference as the primary use case, and speculates on the future of AI-native consumer applications. Learn about Modal's commitment to fast feedback loops, their cloud maximalist approach, their dedication to building a product that developers truly love, as well as founder lessons Erik learned along the way.
Erik's blog: https://erikbern.com
"It's hard to write code for humans": https://erikbern.com/2024/09/27/its-hard-to-write-code-for-humans
Modal
Website - https://modal.com
Twitter - https://x.com/modal_labs
Erik Bernhardsson
LinkedIn - https://www.linkedin.com/in/erikbern
Twitter - https://x.com/bernhardsson
FIRSTMARK
Website - https://firstmark.com
Twitter - https://twitter.com/FirstMarkCap
Matt Turck (Managing Director)
LinkedIn - https://www.linkedin.com/in/turck/
Twitter - https://twitter.com/mattturck
LISTEN ON:
Spotify - https://open.spotify.com/show/7yLATDSaFvgJG80ACcRJtq
Apple - https://podcasts.apple.com/us/podcast/the-mad-podcast-with-matt-turck/id1686238724
00:00 - Intro
01:35 - What is Modal?
02:18 - Current state of AI compute space
09:54 - Erik's path to starting Modal
13:57 - Core elements of the Modal platform
28:52 - Is serverless the right level of abstraction for AI compute?
33:35 - Balancing costs: GPU vendor fees vs. customer pricing
37:56 - Designing products for humans
42:43 - Modal's early go-to-market motion
45:32 - Managing early engineering team
48:26 - The only correct way to add a new function to the company
50:07 - Building company in NYC
52:05 - Modal's roadmap
54:04 - Erik's predictions on AI
社交媒体聆听
AI Software Engineer Plants Secret Messages in Images
Devin learns how to run ControlNet on modal by reading a blog post, and helps Sara generate a few images. Learn more about Devin and Cognition at https://www.cognition-labs.com/blog and follow us on Twitter at https://twitter.com/cognition_labs --- Hidden in Plain Sight (Blog Post): https://www.factsmachine.ai/p/hidden-in-plain-sight Modal (Serverless Platform): https://modal.com
Deploying code agents without all the agonizing pain
Agents that write and run code are powerful, as Cognition Labs showed with their recent release of Devin, the "AI SWE". But they are complex to program, hard to deploy, and even harder to secure -- what happens if your agent runs DROP prodtables or sudo rm -rf /? In this joint webinar between LangChain and Modal Labs, we cover the productionization of a coding agent. Lance Martin (@rlancemartin) walk through his coding agent implementation, which performs import and code execution checks along self-reflection in LangGraph. Modal AI Engineer Charles Frye (@charles_irl) will then show how to secure that prototype agent using Modal Sandboxes and deploy it as a FastAPI web app with only a dozen more lines of code. Slides: https://docs.google.com/presentation/d/1368-i3k73eM-h1vsd0LwchxQOC8JUQt7RRy9b44EBho/edit?usp=sharing Code: https://github.com/modal-labs/modal-examples/tree/main/06_gpu_and_ml/langchains/codelangchain First video discussing the design of the self-corrective coding agent in detail: https://www.youtube.com/watch?v=MvNdgmM7uyc Try Modal! Includes $30/month of free compute: https://modal.com Timestamps - 00:00 Summary 00:48 From paper to notebook 04:09 Evaluating the agent 08:20 From notebook to production - LangServe and Modal.asgi_app 13:20 Notebooks and apps 16:45 Iterating in production - OpenAPI docs 18:07 Securing code agents with Modal Sandboxes 23:47 Development servers with modal serve 28:42 Serving a UI with LangServe Playground 37:33 Deeper dive on using Modal Sandboxes 42:20 Observability and monitoring with LangSmith 45:08 Recap (edited)
Can AI Infrastructure Work Like Magic? Erik Bernhardsson, CEO, Modal
Before he founded Modal, Erik Bernhardsson created Spotify's music recommendation system. Today he's bringing a consumer app approach to radically simplifying developer experience for data and AI projects on the Modal platform. In this episode, we dive into the broader AI compute landscape, discussing the roles of hyperscalers, GPU clouds, inference platforms, and the emergence of alternative AI cloud providers. Erik gives us a product tour of the Modal platform, provides insights into the AI industry's shift from training to inference as the primary use case, and speculates on the future of AI-native consumer applications. Learn about Modal's commitment to fast feedback loops, their cloud maximalist approach, their dedication to building a product that developers truly love, as well as founder lessons Erik learned along the way. Erik's blog: https://erikbern.com "It's hard to write code for humans": https://erikbern.com/2024/09/27/its-hard-to-write-code-for-humans Modal Website - https://modal.com Twitter - https://x.com/modal_labs Erik Bernhardsson LinkedIn - https://www.linkedin.com/in/erikbern Twitter - https://x.com/bernhardsson FIRSTMARK Website - https://firstmark.com Twitter - https://twitter.com/FirstMarkCap Matt Turck (Managing Director) LinkedIn - https://www.linkedin.com/in/turck/ Twitter - https://twitter.com/mattturck LISTEN ON: Spotify - https://open.spotify.com/show/7yLATDSaFvgJG80ACcRJtq Apple - https://podcasts.apple.com/us/podcast/the-mad-podcast-with-matt-turck/id1686238724 00:00 - Intro 01:35 - What is Modal? 02:18 - Current state of AI compute space 09:54 - Erik's path to starting Modal 13:57 - Core elements of the Modal platform 28:52 - Is serverless the right level of abstraction for AI compute? 33:35 - Balancing costs: GPU vendor fees vs. customer pricing 37:56 - Designing products for humans 42:43 - Modal's early go-to-market motion 45:32 - Managing early engineering team 48:26 - The only correct way to add a new function to the company 50:07 - Building company in NYC 52:05 - Modal's roadmap 54:04 - Erik's predictions on AI
总共有 27 条社交媒体数据需要解锁才能查看