cloud-oak

The final example of the DnD statt sheet makes me think whether it's legal to nest <dl>s?

I.e. can we do

    <dl>
      <dt>Actions</dt>
      <dd><dl>...</dl></dd>
    </dl>

You can always ask Codex to fix Claude, issue solved!

> Training AI models can generate enormous carbon emissions

Sure, but what I'd really like to see is a graph for how much carbon is generated serving these models globally.

This one caught me completely off guard when opening YouTube the first time on an iPad: Accidentally clicked on a wrong button and got stuck in a "please subscribe to premium" modal. No amount of swiping or tapping outside the popup would help, only thing left was killing the entire app.

This experience put a major dent in my perception of the "Apple has the most intuitive UI" narrative.

This is already a problem with things like Mastodon - as soon as you subscribe to some more "spammy" accounts such as news outlets, all the other content is drowned out.

So yes, having kind of re-ranking _algorithm_ can be a good thing, whether we like it or not.

Picking up your original comment - Markov chains are much better at generating random numbers than either humans or LLMs!

What's really worrying is seeing medical professionals starting to rely on these tools.

My wife had a pretty bad cold during pregnancy and our GP proceeded to prescribe her cough syrup with high alcohol content, because that was what ChatGPT told him to prescribe. We only noticed it once she took the first dose and spit it out again...

I think Mattermost lost a lot of instance admins' trust when they recently decided to update the server to limit access to old messages without good reason. On self-hosted instances!

https://github.com/mattermost/mattermost/issues/34271

That's why double-blind review shohld be the norm. It's wild to me that single-blind is still the norm in kost disciplines.

The EU mandates that all large data centres built/commissioned from July this year will make use of waste heat:

https://www.twobirds.com/en/insights/2024/germany/rechenzent...

I know of this one [1], a 1000W space heater with integrated cryptominer. Looks kike you can actually buy it now. Not sure how much the mined crypto offsets the heating costs though.

[1] https://21energy.com/products/ofen-2

Sadly had a very similar experience about the screen of my FP4, which seems to have a serial fault of producing random inputs whenever it so pleases [1]. Knowing I had bought a phone with great self-service claims, I was confident they could just send me a replacement screen and I could swap it myself. But no, they insisted that I had to send it in, claiming that this would be better for the environment.

I do want to support Fairphone's mission and wish I could whole-heartedly recommend it to friends and family. But this experience and the many software issues have led me to recommend other options instead.

1. https://forum.fairphone.com/t/ghost-inputs-on-fp4/82837

My understanding of this was that the UB starts only after the value is passed/returned. So if foo() has a contract to only return positive integers, the code within foo can check and ensure this, but if the calling code does it, the compiler might optimize it away.

I was very reluctant about the polars syntax as well initally, but it has grown a lot on me.

Pandas syntax is super ergonomic for quick one-off analysis, but it becomes hard to read/maintain once your processing gets more complex.

For example, the innocent

    df[arg]

can mean wildly different things - does it filter rows? Subset columns? Extract a single column as a pd.Series? There really is no way of knowing except for checking the value of arg.

In contrast, polars syntax feals clunky initially, but it's much easier for me to revisit a pipeline and quickly understand what it does.

SQL more elegantly introduces ternary logic in this case, where any comparison with NULL is itself NULL. This is sadly not possible in most languages where a comparison operator must always return a (non-nullable) boolean value.

It is implemented using web maps technology (https://leafletjs.com/), similar to e.g. the Google Maps satellite view. The screenshots are then served pre-assembled into quadratic map tiles at different zoom levels. This way the client only ever has to load and display a hand ful of relevant tiles.

Perfect analogy! The cool part is that your model also gives good intuition about the gradient descent part. The springs' forces are the gradients, and the act of the line "snapping" into place is the gradient descent process.

Technically, physical springs will also have momentum and overshoot/oscillate. But even this is something that is used in practice, gradient descent with momentumg.

With some notable exceptions. I'll never love the turbofish [1] for example.

[1] https://github.com/rust-lang/rust/blob/master/tests/ui/parse...

The funny thing is that the original Desktop UI is still in the Android version. Sometimes I'll get jokers/consumables stuck in the desktop's "selected" mode, where the tiny sell/use buttons pop out underneath. So the whole drag-n-drop thing seems to have been added on top of the original ui, and sometimes the old ui events still fire.

Example: https://i.postimg.cc/P5VrVhQ4/Screenshot-20250101-103921.png

I thought it was only me! This "inverse question phrasing" is something I see popping up more and more in forums and such, especially phrases starting with "How it <verb>?" or "Why it <verb>?".

The thing is, 3D to 2D projections work well because that's how our eyes work in the first place, so we can intuitively understand a projected image on the screen.

When projecting 4D to 2D, weird things start happening. Instead of a ray, each pixel now represents a plane in 4D space. So even while it can be done mathematically, the brain is just overwhelmed by what happens when you e.g. rotate the projection