Veo's Viral Videos, Upgraded Agents and Dashboards in 10 Minutes

Google’s Veo stuns the internet, OpenAI upgrades its operator and Perplexity Labs mode goes full analyst.

Fellow Data Tinkerers!

It’s time for another weekly round-up on all things AI and data. But before that, I want to mention again, if you are a data person in the trenches and interested in sharing your experience and learnings with other people, reply to the email or message me and we can work it out. We get almost 10k views every month and your experience would be valuable to others

Now, with that out of the way, let’s get to this week’s round-up!

---

The Buzz 🐝

• Google’s Veo 3 is still lighting the internet with realistic videos. Here are a few interesting ones I came across. From a ‘what if Jurassic park was real’ video:

To a “pharmaceutical ad” for a puppy for hire service:

Most of the videos have been pretty short (less than 1 min) but how long before we see realistic movie-length videos? A type of question movie studios are definitely thinking about as well.
If you still want more, the video below has compiled some of the most viral ones.

• Moving on from Google, OpenAI is upgrading Operator (their AI Agent) to run on its o3 reasoning model, promising stronger math, reasoning and safer web-browsing skills. Another agent related news, Hugging Face launched a free, cloud-based AI agent called Open Computer Agent that can use a virtual machine to complete tasks. The agent might be slow or sometimes stumbles on complex jobs

• And last but not least, Perplexity Labs is now available to Pro subscribers. It adds features handling complex tasks like building data dashboards in up to 10 minutes per project. Unlike the quick Research mode, Labs digs deeper and delivers richer, more interactive results.

---

Data Science & AI

• 20 Pandas One-Liners That Can Save You Hours of Work
  

  Cornellius Yudha Wijaya
  discusses 20 Pandas one-liners that’ll save you hours of data-wrangling pain. Think faster filtering, memory fixes and no-nonsense tricks for your next Python project.

  

• Highlights from the Claude 4 system prompt
  This is an interesting analysis by

  Simon Willison
  of the Claude 4 system prompt which was released 10 days ago. It instructs Claude to actively fact-check users and includes hardcoded election results to counter training data confusion.

• How Reddit Scans 1M+ Images a Day to Flag NSFW Content Using Deep Learning
  Reddit needed to flag NSFW images the second they were uploaded. They built a deep learning system that does exactly that; fast, scalable and battle-tested in prod. Here’s how it works.

---

Data Engineering

• DuckLake: SQL as a Lakehouse Format
  DuckDB dropped DuckLake, a new lakehouse table format that stores all metadata in a regular SQL database instead of scattered JSON or Avro files while keeping data itself in open formats like Parquet. This design simplifies metadata management, improves reliability and supports features like ACID transactions, schema evolution and time travel.

• How Notion Brought Order to Its Data Chaos
  If you want to know how Notion’s data went from total chaos, wild JSON, missing docs, nobody knowing what anything meant to a system where most events, tables and definitions are actually discoverable (and up to date), check out this article

• Boring Iceberg Catalog — 1 JSON file. 0 Setup
  

  Julien Hurault
  just built the “boring” Iceberg catalog: no servers, no APIs, no config headaches. Just a single JSON file and a smooth CLI to spin up Iceberg tables in seconds. So if you have had it with Iceberg catalogs, definitely check this out!

---

Data Analysis and Visualisation

• How to Reduce Your Power BI Model Size by 90%
  A detailed breakdown of how Power BI’s VertiPaq engine stores and compresses data, showing that most model bloat comes from unnecessary columns and high-cardinality fields. By trimming unused columns and optimizing granularity, the author cut a real-life Power BI model from 776 MB down to 90% smaller.

---

"Claude 4 refactoring"

(Source: Twitter)

---

If you’re interested in doing a guest post, just reply to this email or message me on Substack and we’ll tee something up.

---

Keep reading: