At the SMB scale accountants are mostly paid to coach/pester/goade the employees to hand in the necessary paperwork in time. The accountants job is relatively quick from there on.

Anytime algorithms are great for robotics planning, for example. A plan does not have to be perfect to be useful, especially when it can be refined further in the next timestep. And the robot cannot act out the plan instantaneously, so by the time one is close to the point where a non-ideal segment would be, one has had many timesteps to refine/optimize it. But robot could start moving right away.

Grid Style Sheets / GSS was an implementation of this idea. https://gss.github.io/guides/ccss

I saw was because I do not think anyone has used or maintained it since the company behind went belly up. But the code is still out there as open source, probably can be learned from.

Disclaimer: I used to contract for them, but in other areas.

If I am to provide a summary of why layout shouldnt be linear constraints, is that it can't faithfully represent content overflowing onto multiple lines. I.e. it's inherently one dimensional. I.e. you can't really have a layout that adaprts to the screen size without creating a lot of separate breakpoints. This is a big limitation, that for example flexbox doesnt have. When I left Grid i immediately went and reproduced a lot of stuff we've been doing in (new then) flexbox layout engine, and i was like: Oh my god, this is so much more powerful.

Good point about overflow. Did you ever do any write-ups about your flexbox experiments, or constraints for layout more generally?

Hello Jon! This is Yarik. Cool to see you doing all this machine learning work these days.

Hi Yarik! Nice to hear from you again. Hope all is well :)

Yeah the USB stick enables the participants to replicate it more easily at home or with friends etc. Encouraging that the participants are in the driver seat also helps with this.

And sharing/spreading the knowledge widely. If 1000x are doing something smart vs a few here and there, makes a huge difference in a marketplace.

Or domesticate them. A superintelligence could probably figure out how to do that with humans, _if_ they saw a reason for it.

High performing types in those areas network a lot, and just as much off the clock (or more) as on. From those I know, many of them seem to pick hobbies, friends, holidays, activities, gym, etc that is compatible with connecting with the right people. And that is considered very normal, more ore less expected if aiming for partner type roles.

The pandas workflows have also been stable for the last decade. That there is a new kid on the block (polars) does not make the existing stuff any less stable. And one can just continue writing pandas for the next decade too.

Code using pandas is testable and reusable in much the same way as any other code, make functions that take and return data.

That said, the polars/narwals style API is better than pandas API for sure. More readable and composable, simpler (no index) and a bit less weird overall.

Polars made the mistake of not maintaining row order for all operations, via the False-by-default argument of maintain_order. This is basically the billion-dollar null mistake for data frames.

Yeah that really should have been default. Very big footgun, especially when preserving ordering is default in pandas, numpy, etc. And especially when there is no ingrained index concept in polars, people might very well forget that one needs to have some natural keys and not rely on ordering. One needs to bring more of an SQL mindset.