𝔸𝕝𝕝𝕠𝕣𝕚𝕄𝔻

We will have papers written by AI, peer review done by AI, and then no one will read them anyway because they will just ask AI to summarize the paper for them. Complete circle!

I always enjoyed the episodes where you interviewed developers about the behind-the-scenes architectural design decisions that they had to make during iterative development of, say, SQLAlchemy, FastAPI, Pydantic, etc. I’d love such stories — they highlight the ingenuity and perseverance of devs.

Maybe some PEPs attempt to correct some peculiarities, but maybe some persist. What would people change if they could?

Here’s a fun one: In the spirit of “What if…”, have a panel of experienced Pythonistas sharing aspects of the language that they wish were different. For example, maybe early design decisions made sense but ended up backing things into a corner of technological debt that has been hard to undue.

It might be interesting to hear about Mojo, the new language being developed by Chris Lattner. It is supposed to be similar to Python. But what problems does it attempt to solve?

I like bread ‘n butter computer science stuff. Anything related to algorithms, databases, etc.

I appreciate the update on your YouTube channel about this math problem. It goes to show that LLMs are aimed at language, not at math... even though agentic approaches may change that situation in the future. We need an LLM-to-WolframAlpha type of agent.

ChatGPT is actually pretty good at rendering quasi-XKCD comics!

On a side note, I have often wished that macOS' symbol character viewer app were better!

That is, a key while in a Quarto .qmd document, and it inserts the LaTeX code. Press the same key in Word, and it inserts the unicode glyph. For the user, it is simply the same hardware key. The software would detect the context and choose the output.

Hmm. Perhaps a bit much... *but* it makes lots of sense if one needs to type the same symbols in various apps. Then the same physical key could be linked (in keyboard settings or via BetterTouchTool) to output the corresponding code for LaTeX or HTML or the unicode glyph itself for a word processor.

Yes, I've been incorporating Markdown conversions in my own app, and it has many inconsistencies/variations and edge cases. Even whitespace control is tricky.

Whoa! Mind blown! 🤯

This looks very promising.

Thanks for sharing this. I never heard of `just`, but it may be a very useful tool in the future. I have mainly maintained my own bash aliases for this purpose, but justfiles can keep things project-focused and out of my .bash_profile.

400+ attendees! That's amazing!! Congrats on a successful summer course.

It's amazing to see the tremendous growth of Marimo. And @koaning.bsky.social must be in heaven getting to develop cool features and showing off cool party tricks. Thanks for working so hard to improve notebooks!

This video was very inspiring. Thanks for taking the time to share it. You're really creative... It got me thinking about making cool "middleware" for some of my own projects.

I figured out how to do this using a combination of bash functions plus `direnv`. As a plus, my custom bash prompt displays the virtual environment name and the Python version. Since it works via .bash_profile, it even works in PyCharm's terminal!

cd and virtualenv This feels like a fun thing to add to your `.zshrc` file. ```bash cd() { builtin cd

@koaning.bsky.social Hi! I saw your post at koaning.io/posts/cd-and.... I had done something similar — but since half of my projects still use `conda`, while new projects use `uv`, when I cd into the directory, I had to figure out which environment to activate.

I hope you get well soon.

@koaning.bsky.social I read your post about command-shift-C in the browser to select an element and bring it up in the DOM. It works in Safari, too! I had no idea! Thanks for sharing this great tip.

Textual - Home Textual is a TUI framework for Python, inspired by modern web development.

Textual is the best! Love it. textual.textualize.io

Great episode! I learned a lot. I like these deeper dives into the core functionality of Python. I can't wait to try t-strings. I have a project that uses Jinja, but I'm tempted to remove the dependency and use t-strings instead.

You are so creative.

Hello! Good comments. Also, it is a terrible practice to have two Y-axes (one on the left for one plot, one on the right for the other plot). The shape of the plots is dependent on the scales chosen, and the "intersection" means nothing. Instead, one should use two separate charts.

I see you used Pearl's `do()` operator in the last line. (The error was in the lines above, where `==` in your DSL was translated to `do()` in the LaTeX output.)

Very creative. Note that `P(Y | X==x)` is not the same thing as `P(Y | do(X=x)`. In your LaTeX line, the equivalence operator is translated to a `do()` function. You could require Pearl's `do()` operator, or use the walrus operator notation `P(Y | X:=x)`, to explicitly mean treatment assignment.

If you are thinking of extending this to allow for do-calculus in causal inference, I recommend using the walrus operator to designate variable assignment (`X := x`), in contrast to the equivalence testing that you show (`X == x`). I have found the walrus operator valuable for providing clarity.

And if you had wanted to condition on the union of conditions (typically using `|` for `OR`) instead of the intersection of conditions (`&` for `AND`), how would you have managed since the initial pipe meant "conditioned on"?