Skip to content

Denny Lee

Databricks

  • 1 episode on Chain of Thought
  • 6 topics covered
  • 1 full transcript
  • May 2025 first on the show
Denny Lee

At the time of the conversation, Denny Lee was Director of Developer Relations and a product management leader at Databricks, and a longtime contributor to Apache Spark and MLflow. On Chain of Thought he connected AI’s foundational data work to its frontier applications, explaining why logging, tracing, and data lineage are needed for credible evaluation.

Cover art for AI's Two Extremes – Foundations & The Frontier | Databricks’ Denny Lee

Latest episode with Denny Lee · EP 22

AI's Two Extremes – Foundations & The Frontier | Databricks’ Denny Lee

Denny Lee, Databricks ·

Show notes & transcript → All listening options →

Episodes

What the conversations cover

These notes summarize the episodes featuring Denny Lee, including other speakers in panel conversations. Follow a source link to read the discussion in context.

EP 22 AI's Two Extremes – Foundations & The Frontier | Databricks’ Denny Lee

  • Logging and tracing feel boring, but they’re what everything else rests on. Denny Lee traces MLflow’s origin to a familiar mess: hyperparameters in spreadsheets, versions named “golden,” “golden_v1,” “golden_v2,” until nobody recalls which set built a given model. Across millions of automated iterations, you can only evaluate what you recorded. Episode takeaways
  • The fight isn’t for one standard — it’s against having fifteen. The industry needs a shared way to record what matters — hyperparameters, lineage, architecture — so any tool can read it and you keep your choice of tooling. Lee doesn’t care if it’s MLflow, OpenTelemetry, or something else: “We just want one,” since each new unifying standard just becomes number sixteen. Episode takeaways
  • Synthetic data isn’t fake data — treating it that way gets the math wrong. Lee’s correction: it’s generated from real data to preserve the patterns a model needs while protecting privacy. If it were truly fake, it would carry a different pattern and different math. Either way the ceiling is the real data — “you’re only as good as your data.” Episode takeaways

In their words

Just as long I have the tooling and I have the data, then I can actually go ahead and improve the model. And then that becomes your IP. And so that's actually how you reduce the cost because from where we're sitting from and also from Galileo's perspective, right, the reality is we don't want you to go ahead and pre train unless you need to. Don't get me wrong.

Denny Lee AI's Two Extremes – Foundations & The Frontier | Databricks’ Denny Lee · 18:28

Transcript excerpts; ellipses mark omitted words. Some automatically selected quotes have filler words removed.