I spent 8 hours debugging datetime parsing in a PySpark migration. Every solution failed… built-in functions returned silent NULLs, UDFs took 6 hours to run, pandas UDFs still failed on 8% of rows.

Sorry, put the wrong link. I wanted to post to the blog. This is hte correct one

https://www.datacompose.io/blog/introducing-datacompose

Why are they using Kendrick Lamar songs for each title?

It's honestly really weird and makes reading the post hard for me.

This isn't really meant to replace therapy to be honest, it is more meant to be a stop gap between appointments for therapy if that makes sense.

A note: This doesn't work great in mobile, or in firefox/safari browsers due to the expiremental files used

Thank you!

Though if you have clean code… it makes it easier to iterate

Tinygrad in shambles

How?

9 bucks per developer isn’t that much. I getting a developer to be 5 percent faster is a huge gain for just 9 dollars.

Sure, but the harder part is measuring whether the developer is actually 5% faster. Otherwise you can make the same case for every $10/mo subscription service in the world, and so we should all be operating at infinite efficiency.

> Sure, but the harder part is measuring whether the developer is actually 5% faster. Otherwise you can make the same case for every $10/mo subscription service in the world, and so we should all be operating at infinite efficiency.

After n such iterations, the developer gets 100*(1-0.95^n)% faster. So, after some such $10/month purchases, the developer gets so fast that buying another improvement yields diminishing returns.

Certainly for SV salaries. But $19/month per developer in addition to already existing cost can make a difference for regions with lower salaries.

I agree. I can see these AI assistants becoming a game changer and eventually a requirement to keeping up, but the costs will be prohibitive for engineers in many regions.

That’s gonna be nice until you have to do a UDF on a data frame