Looks like an amazing role 👀

This is sad — maybe for a different reason than you might think. There are very few ways to measure whether or not a change has an effect. The most consistent one is to randomly provide the change to some and not others. What I’ve seen over and over, is to folks responsible for

I think it's already been done. It's called Hex. If only it were more snappy.

Love this engagement from the always thoughtful Ibis team.

The cost of software isn’t just the development of the first versions. It’s the upkeep and the modifications. I’m long devops / mlops

using pyspark for the first time since 2021, and oh man has it gotten wayyyyyy faster (and simpler to get started)!

Community Notes wouldn’t work well without negative rating signal. But you have to be smart about how you use them. If you naively add them all up, you’ll get a hivemind like Reddit. One way: only downrank if you see negative ratings from people who typically disagree

stop using pandas what are you doing my god?!

unless you're doing time series stuff (maybe), just stop using pandas

1. write the code. 2. write the design doc and share. 3. update the code based on suggestions. 4. submit pr & profit

Here are a few reasons why I prefer using Ibis over Pandas/polars data ( bdw I still use Polars or Data Fusion as the engine ) - Ibis is lightweight and directly executes queries on the underlying databases, functioning more as a result transporter than a heavy framework. -

I remember speak to Hex product people as a customer about the inefficiency of the pandas memory structure, and they all sighed and said "we know!". Glad to see lots of work coming together to make the product even better!

This. Finally, this. ❤️ ``` spark = DatabricksSession.builder.serverless(True).getOrCreate() ```

Love this contradiction. Both Noah Smith 🐇🇺🇸🇺🇦🇹🇼 and the earnest replies are correct. Modi is less popular in South India, but more popular among techies (who skew south indian) than lawyers/policy types (who skew north indian). Indians in the US are a very biased sample.

I should stop sneering at “spreading” in American debating. Given the utility of voice to text, spreading might become the speed typing of tomorrow.

While this conversation was fantastic on so many levels, I can’t help but wonder — is training vs exploration just the same debate as correlation vs causation being played out at a grander scale?