This is also a battle of someone who has historically taken corporate PAC money vs someone who hasn’t

Righteous fury doesn’t exist for those taking big money (and I don’t trust that swearing off it going forward is good enough when there are other options)

Theres also environment and package restrictions, for instance pandas comes pre installed and you’re forced to use certain releases. py312 still wasn’t supported last I checked

It’s also just hard to reproduce a notebook that switches back and forth between Python and SQL-on-pandas

In notebooks edit mode, truly everything is saved as opposed to having commits

I’ve seen large disparities between the published version and the edit mode version that become too difficult to reconcile over time.

Same, I had been wanting to try it for years too. At this point I’d be more interested in self hosting Marimo notebooks and writing everything myself if I had the choice

If you’re referring to hex notebooks, I gave up using their visualizations for that reason and just plot in base python in the notebooks.

Between that and their environment and version management, I’m not their biggest fan - it’s clunky

The over regulation is already happening in cities like NYC and Toronto, not sure why you’re assuming Im not pay attention.

Cycling infra solves for these problems I mentioned, but we also need more clarity about sidewalk usage or else the over-regulation is going to become more extreme

I’m not shocked, cars are a leading cause of death. I’m as anti car as it gets

This isn’t a one thing or the other situation. Make the e-bike/escooter situation better to prevent future backlash and negative talking points. Everyone wins

All for it, but in many US towns and cities some are using e-bikes and escooters recklessly - there definitely need to be better guidelines. I’ve experienced it myself and even know someone who was hospitalized from an escooter hit and run

Everyone does this everywhere sadly

94% is popularized from @rmcelreath.bsky.social and subsequently ArviZ for being as arbitrary as 95% - not sure if you’re implying it’s p-hacking but I wouldn’t rush to conclusions if so.

Stuff like this is why Pre-registration is so important and should get more popularized

This is great, I’ll definitely be sending this peoples way when it comes up at work

I don’t have a paper in mind, but I simulated a really simple example here why VIF can be overrated kylejcaron.github.io/posts/2025-0...

Imo everything goes back to causality, VIF and multicollinearity aren’t nearly as important as having a valid DAG for your data generating process

Is there anything big you expect to get blocked?

The stuff you list is great but when ICE is now so funded and judges are losing their power, I’m spiraling

Could you elaborate or give some examples to make me feel better

Am I crazy or is the description off? Just skimmed through it and it seems like it’s more of AR models, GAMs, and GPs

Yeah I could imagine that landlords wouldn’t be happy, the “fair compensation” would have to be very clearly defined. And future sales and investment would be limited due to downward pressure from lower rents and affordable housing.

Sounds great to me but somehow this is the country I live in

Not trying to be skeptical, just trying to understand the specifics. Anytime I bring this up in conversation, that will be the talking point so it’d be nice to have an answer ready

How do big cities get started on this? I saw the whole first dibs government policy, but I could picture alot of landlords becoming upset there isn’t competition driving the sale price of their investment property

He’s a better coach than he gets credit for

How Americans react to their freedoms being stripped away: 📈

Any good reads to point me towards in this area? I hear people talk about this stuff all the time now…

I’m still fairly new to VI but learning more in my spare time - i do more applied statistics but the idea of scaling Bayesian models and still getting robust uncertainty would be amazing

Exciting to see all of these branches of new VI work and looking forward to reading through your paper more!

Referring to this stream of work from Agrawal, I’ve had it in my backlog for a few months but I saw a blog post from Bob Carpenter talking about alot of success with flow VI in terms of good uncertainty quantification and scalability compared to MCMC

scholar.google.com/citations?vi...

Nice work! Still reading through but what benefits do you see for this method over other approaches like flowVI and eigenVI?

VI tends to underestimate uncertainty, does this suffer from that same problem or does it do any better in that regard?

This is a very underrated answer. Always found these very helpful for calibration plots in low signal to noise problems

As a Milan fan that watches every game, things were going fine for him until recently. He was getting good minutes and putting in good shifts, but his decision making and composure aren’t there yet and he probably got in his own head.

A red card and 2 mistakes leading to goals later and here we are

It’s definitely a culture here. People grew up reading car magazines, view it as a symbol for freedom, and completely disassociate from reality and view traffic not as a car problem but as a city problem.

It’s insanity, it’s why cars are a leading cause of death, and it’s partly why we cant change

It should be noted that raising taxes above 100% is proof that the administration *does* understand that the tariff is *not* paid by the foreign country & seller (who would not sell if he had to pay >100%) but by domestic consumers

Really recommend learning uv or poetry like @lemonwatcher.com brought up. Doesn’t take too long to learn and solves for all of this