I have been leaning the other way. There’s room for nuance in the discussion but a stance of certainty that full remote is just more effective screams “expert beginner.”

There’s a third option: some people work best alone / remote. Some people work best in an office.

There’s more than 3 options. That’s what the nuance is about.

Sometimes individual productivity isn’t even the measure of success and sometimes it needs to be sacrificed for group productivity.

> some people work best alone / remote. Some people work best in an office.

this is true, and irrelevant to my reasoning.

This idea that if you change your app in a way that some people don’t like, then the government legislate your features for you, is just mind boggling to me.

I think this really only becomes an issue if the changes you made to your app is in some way affecting democratic elections in that country (and in that case, seems well within the purview of legislation)

Not if there’s enough of them.

Unless you depend on Facebook for your livelihood, then this is a false equivalence.

RSU grants assume a growth rate (15%? I forget) so if they stay flat, go down, or grow slower than the baked-in growth rate, then you make less each year. If you do well enough, they’ll give you some RSUs to “make you whole” (as they used to say) but that doesn’t really happen anymore (or not much).

This is not true for your initial four year grant. I’m going to make up a number to make the math easy. Say my total compensation target was $200K. My initial 4 year offer was structured based on the then current stock price.

It would have been what ever it takes where base + prorated signing bonus + RSUs would equal $200K taking into account the 5/15/40/40 RSU schedule.

The former. Basically: build, train, test, deploy, monitor, repeat for ML algos.

I’ve heard multiple recruiters (from different agencies in different geos) refer to them as “amholes” and said they’re hard to place and difficult to break their bad habits.

I’m sure you know this but it took me entirely too long to really understand and feel the satisfaction of taking and printing my own photos.

It’s very accessible these days to have a finished piece of art that’s all yours - even with little artistic ability.

Are there decent at home photos printers nowadays? I was always underwhelmed with the old inkjet soggy paper prints.

Yup! Both epson and canon have some great printers for high quality printing at home. First Man Photography on YouTube has videos on the topic that I enjoyed.

The surprise for me was that paper quality makes much of the difference.

I’ve used Windows at work for years, my personal/gaming machine is Linux (mint), my personal/development machine is MacOS.

They’re all perfectly viable options with strengths and weaknesses. None of them are especially great. I’m partial to MacOS, personally.

It’s willful ignorance to think that the many millions of people that like MacOS are just parroting what they’ve been told.

> It’s willful ignorance to think that the many millions of people that like MacOS are just parroting what they’ve been told.

This is so entirely true.

I've installed so many different Linux distributions (and multiple Windows versions) on my personal laptop. Currently noodling around with NixOS.

I've never been tempted by a non-macOS laptop for work.

Whatever faults macOS has, it is very good at staying out of my way for getting work done, and all the small ancillary bits (eg webcam and audio support for chatting) have worked flawlessly for me for two decades. I cannot say the same about either Windows or Linux.

Take whatever you're indexing and make it 16-20x and that’s a good approximation of what the vector db’s total size is going to be.

Why is it like that, currently? There is no information added by a vector index compared to the original text. And the text is highly redundant and compressible with even lossless functions. Furthermore a vector index is already lossy and approximate. So conceptually it is at least possible to have an index that would be a fraction of the size of what is indexed?

There is some information added, depending on the vector db and context (some systems will add permissions related metadata so that the LLM won’t pull chunks that the user didn’t have access to).

The vector itself is pretty large (512 dimensions).

The chunks have an overlap (iirc 30% but someone feel free to correct me).

I don’t _think_ the data is typically compressed (not sure why but I assume performance).