eximius

By a similar argument, most frameworks are the same, most JS package upgrades are the same, etc, etc.

It's still small incompatibilities adding up. It's still work.

React today looks very different from React from 10 years ago. Easily as different as being a different framework. So, there is some nuance, but the logic isn't dead quite yet.

I guess you're saying that non-inference time training can be that "sleep period"?

Self updating requires learning to learn, which I'm not sure we know how to do.

Heh, no chance I can introduce this at work and hard to have a personal project requiring it. :)

I think you probably need some documentation to the effect of the current state of affairs, as well as prescriptions as to how to work around it. _Most_ live workloads, even if the total dataset is huge, have a pretty small working set. So limiting DB operations to simple fetches and doing any complex operations in memory is viable, but should be prescribed as the solution or people will consider it's omission as a fault instead of a choice.

I believe that so long as weights are fixed at inference time, we'll be at a dead end.

Will Titans be sufficiently "neuroplastic" to escape that? Maybe, I'm not sure.

Ultimately, I think an architecture around "looping" where the model outputs are both some form of "self update" and "optional actionality" such that interacting with the model is more "sampling from a thought space" will be required.

Something I don't see in the pgdog documentation is how cross-shard joins work. Okay, if I do a simple `select * from users order by id`, you'll in-memory order the combined results for me. But if I have group by and aggregations and such? Will it resolve that correctly?

Glad to see Smithay has approached a point that lots of things are being built on it. Last I really deeply looked at all this was when Way Cooler abandoned wlroots-rs.

And to my pleasant surprise, it seems like there may finally be an AwesomeWM alternative for Wayland now! (Pinnacle)

Ah yes, China is famously averse to threats to the wellbeing of its citizens. /s

When we first got one of our dogs off the street, she was incredibly food motivated. She would hop the fence and roam and bring back cabbages, avocados, cauliflower, chips, and anything else she could find to snack on. Those are just examples we witnessed. But she always came back before we got home from work.

Now, 7 years later, she's picky. She'll spit out some food I give her if it isn't appealing enough. She can still obviously be motivated with food and treats, but it's far less unconditional.

Articulated but not substantiated.

> so we can likely just train only on properly sourced data in some way

But are we? I _can_ source chocolate and coffee from sources that don't use slave labor, but it becomes very important whether you _are_. You can't just shove the moral implications under the rug because an ethical option _exists_. Obviously this is less extreme than my chocolate/coffee example, but considering this is kinda the point of the questions being asked.

3 & 4 & 5 you just completely disregard as issues?!

"3 is not a problem given we do <highly controversial X>" - but what if we don't?

"4 is not a problem because we'll just get better" - but what about now when it isn't and what if we don't?

"5 is not a problem - just make X available and be rich"

> spent quite a lot of time using wire for DI in go only to really study the code it was generating and realizing it truly is code I would normally just write myself.

Yes, but the point is 1) you don't have to write it yourself and 2) it does 'the right thing' even if your junior dev straight out of BS CS wouldn't know how to write it.

There are, of course, caveats and not all frameworks are created equal and any tool can be misused. But I find DI very useful. Personally I'd recommend Uber's Fx framework and underlying dig library.

Others are addressing your point about governments buying data from corporations also being bad.

But also, you think companies like Twitter, Facebook, etc which are increasingly activist and distorting truth and public discourse aren't also privacy threats?

And there is danger of it getting worse. So, your points have merit, but we cannot dismiss the threat of abusive corporations either.

Oh, then the language primer is also out of date.

I do think that the `SOA` struct modifier keyword is maybe the most interesting thing out of Jai. Other languages would do well to think about adopting a mechanism to support something like it.

You can do everything you're saying in Rust.

1. https://docs.rs/dimensioned/latest/dimensioned/

2. https://doc.rust-lang.org/std/ops/trait.Index.html

3. Maybe const generics? e.g., Foo<N: 3> where Foo { array: [u32; N] } or something. You could also do this with less typed-ness with Foo::new_with_capacity(n) and some private size management.

It definitely takes aesthetically from the C-family of languages in the same way that Java and Go are all braces and semicolon languages. I can't exactly minimize this - because I am comfortable with languages that look like this and less comfortable with some languages that don't, so I recognize the effect is very real - but it doesn't affect the features of the language much.

You can already do this.

It's a bad idea because of the state of the rest of healthcare _due to the state of health insurance_, but nothing prevents you from self-insuring.

But were any _language_ decisions discarded due to compile time concerns? I don't think anyone would claim the folks working on the rust compiler don't care.

On that note, thank you for your part! I sure enjoy your work! :)

NPM and pip hell come about for several reasons, one of the biggest being that package versions are global.

In rust, you can project A can use dependencies B and C which can both depend on different versions of D. Cargo/crates generally also solve some of the other metadata problems Python has.

This means the developer experience is _significantly_ improved, at a potential cost of larger binaries. In practice, projects seem to have sufficiently liberal bounds that duplication isn't an issue.

We have multi-room, so I use the Sonos app for EQ between speakers. Sometimes for starting a station from Pandora or playlist from Spotify.

If we need more control over the music, we go to the Pandora or Spotify apps themselves, but Sonos for the speakers.

I guess I didn't use the old one so I don't know what I'm supposedly missing.

It has been bizarre watching this debacle from the sidelines. I must have bought just on the recent side of whatever demarcation of problems are being complained about.

My app works fine. My speakers work fine. The worst problem we have is rarely the music pauses, we're unsure why, and we just hit play again.

I'm wondering if there is some functionality I am not aware I'm missing or if it is limited to older models that I am simply not experiencing the problems with?

Interesting!

Something neat I found, you're able to 'clip' the blocks into each other by spinning them right before the block settles.

I have successfully argued this for at least a certain class of devices. Namely, it practically can't happen _for small devices_. e.g., a TV remote with a voice feature activated by a button simply won't have enough _power_ to constantly be listening, uploading, or processing.

Harder to convince folks that _nothing_ does this, but takes an edge off their more paranoid tendencies.

hah! Clicking "Remind Me Later" for the update prompts works again!

You sound like someone who has been in constant mania from a panic attack for, like, months.

Stop. Breathe. Do something to relax.

You very clearly need it, even if only to center yourself long enough to not throw yourself into the very next thing you see.

My comment is more to address the "whataboutism" than to say the US doesn't have faults.

We exist in a time where nuance is sometimes harmful in public discourse.

The US government is a meandering blob of racoons in a trenchcoat. Occasionally a hand will _do_ evil but even odds whether they realize it or meant it. Partially depends on which particular raccoon is where. Some of the racoons are evil.

China is actively directed by a small group of well coordinated, evil raccoons. See, you know, the constant genocide.

My problem is that I don't want to play sports or doing a running club or any thing like that. And it isn't that I'm not willing to put in some work, it's that I want to make friends who want to do _other things_. I'm not sure how to find good public groups for people like me with the intent of finding friends for less public interests like video games, board games, books, etc.

Is anyone familiar with the (ln B > x) notation being used? What is this value being referenced?