I think it's best to think of this as a good toy example, and not get caught up in the specific utility of a "parent" node.

It is just a bad example.

One of the main points is that there are very few algorithms that require back links that create cycles.

Cyclic data structures, rare as they are, are also hard to reason about. Even for garbage collectors!

If you have to implement one in Rust (I have) it is a bother, a big hassle.

It is a good thing that in the general case when a programmer is implementing back links they are doing the Wrong Thing.

And here I was, innocent, excited to read about a new paradigm for thread management, and wondering why I never knew that wired ran articles that technical.

Now I am cynical and jaded. My disappointment is immeasurable and my day is ruined.

This reminds me of an occurrence when I was around ten to twelve years old and in the shopping mall with my grandmother.

There was a person with a clip board hanging around in the concourse area who asked something to the effect of "I was wondering if you'd like to answer a survey about your genes?"

I excitedly agreed and we were led to a small interview room with a table.

Their questions started out with "What is the brand of pants that you have on?"

I was confused about this bizarre and seemingly orthogonal question as to why we had agreed to be interviewed, and asked my grandmother to look at the tag for me.

It took a few more questions like this before I realized the homophone confusion I had made in understanding the initial solicitation.

I too was hoping for some deep zen of multiprocessing, but then saw the wired.com domain.

I am in way over my head here, so I wasn't able to tell if the authors addressed this, but my intuition is that this should be somewhat mitigated so long as people are providing the filter between which results are discarded and which might end up back in the training pool.

I would think that the human selection process would help to head off this conversion, both by selecting against incorrect results, and also by introducing variance outside of the model. On the other hand since a person can only act as a filter, I can also see how that would be of limited value long term.

They don't address that. They just assume random sampling, so there's no equivalent to human curation or quality metrics, which would preserve tails or, by manual use, create tails. The contraction they observe is pretty much what you would expect in the random sampling setting, since you can only lose tails with a finite sample, and never gain them. (They also need to have a very large ratio of synthetic to real/original data.)

So, while interesting for nailing down that phenomenon, the broader implications everyone wants to draw from it are not very good - very few people are using random GPT-3/4 or Stable Diffusion samples!

We have already heard reports of companies that get paid for human tagging, and similar services, using LLM's to automate their processes.

I used to love fountain pens, they feel amazing to write with, and the variable line thickness makes everything look awesome.

They are also a massive pain, the higher end ones use gold alloys for the nib, making it delicate, if they aren't used enough the ink dries in the nib, they tend to get ink everywhere, and they can be really inconsistent.

Those sound more like dip pens with some old iron gall inks.

Modern nibs are able to be used in pretty much any orientation, and they use modern inks which don't dry out, if you keep a cap on them. I have several pens with caps that sit for years and don't dry out, nor do they require any different writing methodology.

Mercedes, Audi and BMW destroyed their reputation for anyone paying attention a long time ago by producing expensive, unreliable cars that aren't even remotely cost effective to keep running.

Plastic timing chain guides, FFS.

If purposefully building engines that grenade themselves at 80k miles wasn't enough, then it's hard to imagine what would be.

It's really stupid, but you can't tell me that the idea of a manual electric doesn't appeal to the less rational parts of your subconscious.

It doesn't.

Driving ICE manual is the whole package. Engine sound, the manual interaction, the little jerk when you miss the rev on shift, feeling of all the mechanics when changing the gear, feeling the vibrations thru the chassis when you accelerate, engine going braaaaAAAAAP and screaming like angry bees when going past VTEC threshold etc.

And the little details like how the notch on the first gear in my Civic ('07 Type-R) becomes lighter when the gearbox oil heats up and the whole rest of smaller idiosyncrasies of mechanical system.

For me the fun is man understanding and controlling machine, it is gone the moment machine pretends it's something it isn't. Not saying electric wouldn't be fun, just that I hope it will be its own type of fun instead of fake.

My car is a bit raw tho, spins to 8.6k RPM and have pretty hard suspension. Typical ICE car isn't exactly that "talkative" and is much more mushier around the edges.

But it is a bit funny to me that if you want high revving NA engine your options are 15+ years old Hondas or top of the line Porsche...

I’ve driven stick lots of my life and no it doesn’t. You press the pedal on electric and the smooth instant and seemingly endless application of power is one of the absolute best parts of it.

Don’t leave out the old Volkswagens! 7.5k on the tach on my 2.0 16v, very commonly pushed into mid 8s if you end up going to solid lifters and a set of cams.

I agree though, nothing quite like a small 4 banger revving the moon.

It would be like those CVTs with fake shifts: unnecessary, phony, dishonest, and worse than actually embracing the technology and using it properly.

If you want to feel "connected" to a modern car you are out of luck. It won't happen. Adding fakeness to compensate is ultimately unsatisfying.

> you can't tell me that the idea of a manual electric doesn't appeal to the less rational parts of your subconscious.

I can absolutely tell you that. It sounds stupid and burdensome.

And in terms of effectiveness, it's basically homeopathy, or healing candles.

Could we avoid participating in this? The price hikes were pretty brutal, but tesla and their charging infrastructure are still at or near the top of the list when it comes to electric cars that you can actually purchase, at least from the perspective of someone looking to replace their ancient mazda in the next year or so.

Also people don't deserve to be judged by their car unless it is a modern german luxury car in which case they obviously are the kind of person who makes bad life choices.

> Also people don't deserve to be judged by their car

In a way, that was my point, because I was responding to this: "Driving one meant you were forward thinking, that you were on the bleeding edge, and probably conscientious about the environment."

I'm actually really pleased with how well youtube avoids serving you rage bait once you have made it clear to the algorithm you aren't interested.

As a rule, can we avoid dealing in absolutes? Implying that the FDA should have no input whatsoever on laws regarding food safety regulation seems a little odd given that they would represent 90% of the relevant experts.

Who would you rather give a theoretical copy of all of your digital information? Assuming both have equal access to non-digital you.