Marin's hero run is training on our @wandb, and a ten-person open lab is showing everyone how a 535B model actually gets built.
Pull up a chair to the show:
Experiment tracking shouldn’t be a mess.
I explored Weights & Biases (@wandb) with a FashionMNIST Computer Vision project to track, visualize, manage, and analyze ML experiments.
Train → Track → Visualize → Compare 📊
🧵👇
#AI# #MachineLearning# #WandB#
Show more
I built a deterministic city simulation governed by an LLM, which won me the highest prize at CoreWeave Hacks (The Best Loop Design) with Weights & Biases as a solo entry, and they gave me a robot dog.
Fully open source:
@wandb @CoreWeave @typesafeai
Show more
Ship now, fix later. Corey Sanders says that's...mostly right.
Our SVP of Product urges teams to get something out, because production is where you learn.
Follow the rules, keep it secure, then ship. You won't learn if you don't.
Catch the full episode:
Show more
Introducing Cline Desktop - a native interface for working with open weights models.
Use with ClinePass and all our free models like DeepSeek-V4.1-Flash, Musespark-1.3, or BYOK with any provider!
Show more
Moving AI agents from prototype to production faster than ever? Don't let rapid iteration hide critical runtime behavior.
Tune in to What's New Wednesdays: Vibe-code, trace, and deploy your agents to explore:
• Production agent observability workflows in W&B Weave
• Evaluating and self-improving agent loops at scale
• Real-world takeaways from CoreWeave Hacks
Register here:
Show more
We built a simple security layer for agentic infrastructure.
When an attack hits, we don’t let it touch production. We capture it, spin up a sandbox with the same infra pattern, and let the agent learn there.
Same architecture. Different risk level.
The sandbox becomes a training ground: replay the attack → map the failure path → test the fix → store the lesson.
Security for agents can’t just be a wall. Agents write code, call tools, and touch APIs. The next layer has to be a simulator, a teacher, and a memory system.
Attack → sandbox → learn → stronger infra.
Would love thoughts from people building agent runtimes, evals, and AI infra:
@hackgoofer @neutralino1 @JinjingLiang @xdotli @mo_tiwari @typesafeai @wandb @agihouse_org
@ktaletsk @AhmadMustafaAn1 @infiniter3grets @mehular0ra @RebbaVenkatarao
@typesafeai, @benchflow_aiMo @AGIHouseSFAhmad @mehular0ratwo
@shvarts_eugene, @EShvartsman.
@langchain
@LangChainAI
@e2b_dev
@modal
@e2bdev
@swyx
Show more
Thanks @CoreWeave and @wandb! I've been using @wandb since it came out something like 8 years ago. @typesafeai had such blast sponsoring the event.
Super impressed by Christian Jin - who is only 14, who had the coolest typesafe hack (piloting a rocker simulator in real-time!)
Show more
Why not make the Fly compete with you on the dart board. here you go
Compete with the fly here:
@marimo_io @CoreWeave @wandb
Kickoff of day 1 at @CoreWeave Hacks. 200+ builders, one mission: build an agent loop that catches its own mistakes.
@agihouse_org, @typesafeai and @marimo_io are here with $1k+ in credits and tons of compute. 24 hours to submit.
One team is going home with a robot dog!
Show more
Next up Julia showcasing @CoreWeave @wandb ARIA!
Winner for best use of ARIA gets $1,000 as well!
Check out ARIA:
CoreWeave Engineering might have discovered a ✨problem✨
Hey y'all -
Last week I got to speak to @MasterClass Executive Cohort 1! We walked through the @CoreWeave AI Loop, improving agents with @wandb , and covered how to build evaluations for AI systems in a business setting.
Show more
Top 5 of 112 large open-weight models on the @ArtificialAnlys Intelligence Index. Only 18B active parameters.
GLM-5.3-Flash from @Zai_org is live on CoreWeave Serverless Inference: 1M context, vision, $0.15/M in, $0.50/M out.
Ship more for less.
Show more
The research agent (ARIA) in W&B is unreal:
It's great for doing custom analysis and suggesting experiments - it can even create plotly plots in the app
I beat the @wandb intern to it again!
GLM 5.3 Flash is now live on serverless inference on WANDB. 1M Context, vision capable, 0.50$ per million output.
Its also very fast.
@CoreWeave
Show more
What does "Sovereign AI" actually mean for regulated industries?
Join @CoreWeave and @CosineAI on September 10th to learn how to deploy frontier AI securely and how to prove it's compliant.
In a live demo, you'll see how to use W&B Weave to trace every compliance check and red-team probe, helping you build reproducible, evidence-backed approval records instead of relying on self-attestations
Register here:
Show more
A year ago, David was the only FTE on Marin. Today, thanks to Open Athena, Marin has 10 FTE. Read his post to better understand the context of Marin. And David tokens are always a pleasure to read.
DeepSeek-V4-Pro-0813 is live on CoreWeave Serverless Inference.
1.6T parameters. 1M context. Built for long-horizon agent work, and priced for it: cache reads run $0.044/M for all the context your agent re-sends every step.
Start here:
Show more
No more noisy neighbours on your kimi inference
