1) my top-level takeaways on DeepSeek's smallpond: a distributed data processing framework used for training LLMs #databs
03.03.2025 14:57
👍 8
🔁 2
💬 1
📌 0
1) my top-level takeaways on DeepSeek's smallpond: a distributed data processing framework used for training LLMs #databs
One of my favorite things we do on the DX team is when there's a differing viewpoint we each lay out our level of conviction. Often, but not always the lower conviction person will cede to the higher conviction - great sign of a high trust team.
Must read new post from @joellab.es docs.getdbt.com/blog/the-lev...
It's blazing fast!
Thinking of you my friend
Major industry news today: dbt Labs acquires SDF Labs
www.getdbt.com/blog/dbt-lab...
This is, in my opinion, a complete game changer for dbt.
dbt Labs has acquired SDF Labs! Early reviews:
Today, we’re announcing that dbt Labs has acquired SDF Labs. This a monumental day for dbt Labs and the entire dbt Community Thread on what that means 🧵
www.getdbt.com/blog/dbt-lab...
I wish that looks great
Big things cooking at the team meeting today
My contribution: My world famous Karaoke playlist open.spotify.com/playlist/6ss...
It's my birthday! My request - share a song, album or playlist that you love for me to listen to.
My Airbnb has a smart fridge with a memo feature (?) and I couldn’t help myself
This just in - O1 preview coming out in favor of the Oxford Comma. Check the Analytics Engineering Roundup tomorrow to see how the proofreading went and catch some great posts from @faithebear.bsky.social, @rovingrob.bsky.social and more.
Realizing it's a huge miss that I haven't started a newsletter called dbt Blabs yet
Yes!!
Joel you are not going to believe this but he’s been here the whole time
Even got sauce for it
Full house at the NYC dbt Meetup tonight
The findings here based on an RCT:
- Steep increase (39%)! in patent filings and the AI generated ideas rates as more creative
- Gains largely come from the top decile of scientists (81!)
aidantr.github.io/files/AI_inn...
This paper is really worth reading. Shakes up two common tropes that we've been hearing about how AI impacts work:
- That AI is useful for rote tasks but not novel or creative work
- That AI flattens the curve between high and low performers
yes i know it's vendor content but i'm really proud of this blog and worked for many weeks on it, please enjoy and lmk thoughts: docs.getdbt.com/blog/test-sm...
early days still for #iceberg, but you can't ignore a PR that shows me the end of the table format wars is nigh.
Delta-only deletion vectors to land in v3 spec
Incredible to see SWEs from DBRX, SNOW, GCP, and more chime in.
Nature is healing. Iceberg is real! #databs
github.com/apache/icebe...
Aaaand this was the whole response
New record
“I’ve never been part of the Jinja haters league. I’m not part of the Jinja defender squad, but I’m also not part of the Jinja haters league” @joellab.es
Starting off the offsite with my most important piece of thought technology - the mood meter
[guy who used an LLM to write his wedding vow but accidentally used Golden Gate Claude]
“You rolled into my life like the fog rolling into the San Francisco Bay
Interesting bit on how people got mad at the news feed for making it easier to access already public information. 10% of Facebook users joined a group saying to get rid of the News Feed!
Some interesting parallels to LLMs reducing barriers to access already public information
Fun fact - the Facebook News Feed was originally a batch job that ran every 6 hours
The optimal experience while flying is watching Aquaman 2 with no sound while listening to the new episode of Acquired podcast