We're slowing adding features so that you might consider stepping away from powerpoint.
Remember: this can connect to your postgres/bigquery/databricks datasets *and* you have sliders/widgets.
It really took me a while to "get it" when it comes to nbdev. But I gotta hand it to Jeremy Howard this way of working makes too much sense once you're used to it.
As of today, I am working on tools that make this kind of work possible in marimo.
youtu.be/ZLg27UmAJbw
You'd be surprised how often I write emails like this.
I used to run a CRON server locally and I can now happily say that it's been an absolute joy to switch to Prefect instead.
The cli is nice, the dashboard is nice, and the Python code still very much looks like Python. Also super nice to run these jobs locally.
The trick to getting the open source LLMs to be better at Python has little to do with "waiting for a better model". We made a new harness and ... yeah ... I'm using KimiK2 a whole lot more now.
Here's a demo a demo of KimiK2 with marimo-pair:
youtu.be/ZBI7BDUK1Es
This time 'round I didn't review a keyboard but a trackball!
youtu.be/DmV5KP64EwY
People sometimes ask me why I still like to work with Python notebooks given that LLMs can also generate web-apps for us.
The difference to me comes down to the fact that I really like it when my software is usable like legos.
koaning.io/posts/legos...
Yeah there's something general here.
Those who do not have a reputation cannot use it.
It took 6 months to go from 0 to 500k views on YT.
Today we crossed the 1.5M milestone. But it only took us 2.5 months to go from 1.0M to 1.5M views this time around.
Happy to hear it!
Don't trust the video? Feel free to run this yourself locally:
Notebook link:
github.com/koaning/not...
The best April 1st moments are those that aren't merely an announcement, but something that actually works.
So yeah, happy to announce that we got DOOM running inside a @marimo_io notebook!
uv cache prune --force
When you make a lot of demo projects, oh my, make sure you prune UV!
Fun experiment of the day: use a neural network to solve TSP. No gradient? No problem! Evolutionary strategies applies!
Not sure if it is practical yet, but fun times none-the-less.
Have people seen the latest 3b1b video? Because I sure did ...
By the time you see this screenshot, the feature will be live!
Think of it as a new kind of blog where we celebrate Python notebooks :)
... and widgets! So many novel widgets!
This was a fun rabbit hole indeed!
We recently added gcc to molab containers for Cython support, but it turns out that technically ... you could also use Zig! It is available via pypi!
That's an interesting unlock! Feel free to read the nerdsnipe here:
cemrehancavdar.com/2026/03/05/...
Shoutout to alphaXiv, their stack makes it super fun to explore papers!
What's the weirdest prompting trick you know?
Is it repeating the same prompt twice?
No. It's repeating the prompt ... three times.
Details and benchmark in this video: youtu.be/j2viZGPSydY
Been making some dashboards for myself ...
Notebook productivity is going only up now that it is easy to make your own library of JS <-> Python widgets to help you do your work. The interaction is *super* helpful.
Small detail: if you're looking for new DS tools, have you tried marimo?
I find them fastinating, and they are so so useful.
Can be made less scary with hashing tricks sometimes too
It was also interesting to read that you had a similar conclusion. I felt that merely replacing sklearn with a api-compatible rust clone would not be interesting.
... But a tighter ML library on top of polars does offer something new/modern to the DS stack.
I finally found somebody else who thought about porting sklearn! I also spotted the speedup but dropped the ball when I wanted to do sparse vector stuff which is currently no supported in rustland.
Then again, why not try?
Funny you mention that ... that's something we support ;)
www.youtube.com/watch?v=gW3N...
When people start to notice your hobby project ...
yep!
github.com/marimo-team/...
We're getting pretty close to automating stuff like this.
