dennylee

Today at #NVIDIAGTC, we’re introducing AI Runtime, with serverless NVIDIA A10 and H100 GPUs now available on Databricks for training and fine-tuning. GPU infrastructure has been one of the biggest blockers for teams building advanced AI, from long procurement cycles to complex distributed setups and data bottlenecks. AI Runtime removes that friction: - On-demand GPUs with no cluster management - Optimized distributed training with key frameworks pre-installed - Native integration with Lakeflow, MLflow, and Unity Catalog With AI Runtime, teams can focus on building and shipping models, not managing infrastructure. databricks.com/blog/introduci…

Customers are getting some amazing performance with Apache Spark Real-Time Mode! Try it out, just turn a SQL or DataFrame query to a streaming job! This was our-term vision for streaming of Spark since we first designed the API over Spark SQL, and it’s great to see it in action.

"You can think of this as RAG — but like RAG plus plus plus plus plus plus, all the way up to 200 vector database calls." — @jefrankle, Chief AI Scientist at Databricks. We published new research on KARL — Knowledge Agents via Reinforcement Learning — a RAG agent trained simultaneously across six distinct enterprise search behaviors. Most pipelines are built for one and fail silently on the others. KARL generalizes across the kinds of ambiguous, fragmented, real-world queries that enterprise search actually demands. Read more via @VentureBeat: venturebeat.com/data/databrick…

Patrick Ball talks to L’Officiel about his thoughts on Mel and Langdon’s relationship in ‘The Pitt’: “It’s a very unlikely pairing, but it is so funny to me how worked up everybody has gotten over this. I think it's just a beautiful friendship; we don't have enough loving, platonic friendships like that between the genders on television. To me, Mel represents a caretaker—she takes a lot of responsibility for the people she loves, and she knows the weight of it. That is something Langdon feels regularly with his family, too. So when he meets her, he sees that sense of responsibility; he also sees somebody that has their sensitive side intact. This place can be a meat grinder—coming into work and being presented with traumatic situations all day long can create a lot of calluses, and Mel isn't calloused. Langdon doesn't want to become calloused either, and in Mel, he sees this hope that you can stay soft and human.”

Now you can author an AI agent and deploy it on Databricks with Agent Bricks Custom Agents! - Build and iterate locally with your preferred stack - Improve agent quality with integrated evaluation and CI/CD workflows - Deploy directly as serverless Databricks Apps to a consistent, governed production environment - Store and reuse agent state with built-in memory powered by Lakebase databricks.com/blog/custom-ag…

Happy International Women’s Day

Olympic freestyle skiing champion Eileen Gu received a hometown welcome in San Francisco today, serving as Grand Marshal of the city’s annual Chinese New Year Festival and Parade. abcnews.link/Uj9ntz0

A lot of amazing examples, but this is the scene I’m referencing in my tweet!

He failed. And failed again. Then finally nailed it and the crowd exploded.

robby’s refusal to make eye contact or speak because he knows abbot is the only one he can’t lie to and get away with it

Turns out CEOs eating burgers is trending. Honestly… we’ve been doing that since 1954 🤷🏻‍♀️ #burgertime #burgerwars #ceolunch

This Dad recorded his music loving daughter's growth over the years. And his consistent reaction is hilarious 😂 ❤️

The Crosslake Connection opens on March 28. Bellevue Police created a specialty unit - the Bellevue Light Rail Unit (BLU) - specifically for this new era. Our goal: create a safe and positive riding experience for community members wishing to use the 2 Line through Bellevue.

She won gold but she is cheering and comforting a young girl who just got her first medal. Girlhood. Watching alysa liu be the person she probably needed as a kid makes me smile 😊

@NancyMace The only fraudulent activity here is your covering up how you were endangering survivor info, attacking one of the few other survivors in Congress, & refusing to fix serious flaws that endanger victims without consent. Fix it and people will vote YES. x.com/aoc/status/203…

Ukraine offers to help defend the West from Iranian drones. Russia helps Iran track American forces. And the White House eases sanctions on Russian oil. Explain that to the families of U.S. troops. My latest (in reply)

Meet KARL: a faster agent for enterprise knowledge, powered by custom reinforcement learning (now in preview). Enterprise knowledge work isn’t just Q&A. Agents need to search for documents, find facts, cross-reference information, and reason over dozens or hundreds of steps. KARL (Knowledge Agent via Reinforcement Learning) was built to handle this full spectrum of grounded reasoning tasks. The result: frontier-level performance on complex knowledge workloads at a fraction of the cost and latency of leading proprietary models. These advances are already making their way into Agent Bricks, improving how knowledge agents reason over enterprise data. And Databricks customers can apply the same reinforcement learning techniques used to train KARL to build custom agents for their own enterprise use cases. Read the research → databricks.com/sites/default/… Blog: databricks.com/blog/meet-karl…

Awesome job by the @databricks team My summary: They trained a model called KARL that beats Claude 4.6 and GPT 5.2 on enterprise knowledge tasks (searching docs, cross-referencing info, answering questions over internal data), at ~33% lower cost and ~47% lower latency. The key insight: instead of throwing expensive frontier models at enterprise search, you can use reinforcement learning on synthetic data to train a smaller model that's faster, cheaper, AND better at the specific task. RL went beyond making the model more accurate. I t learned to search more efficiently (fewer wasted queries, better knowing when to stop searching and commit to an answer). They're opening this RL pipeline to Databricks customers so they can build their own custom RL-optimized agents for high-volume workloads. I think we'll continue to see data platforms become agent platforms. Databricks' KARL paper is really an agent platform play. The pitch: you already store your enterprise data in the Lakehouse, now Databricks will train a custom RL agent that searches and reasons over it, tuned specifically for your highest-volume workloads (workloads = apps = agents). The business move is closing the loop: data storage → retrieval → custom agent training → serving, all on Databricks. They're turning "your data lives here" into "your agents live here too." Kudos @alighodsi @matei_zaharia @rxin

British Columbia is switching their clocks for the last time this Sunday. There's a lot going on in the world right now, but there is NO reason to not make daylight saving time permanent. Congress should pass my bill to finally lock the clock.

You should follow @j_nadan_chang if you want you keep up with the bleeding edge science in RL.