jacobn

Ah, it's not, will fix, thanks for pointing that out!

Thank you!

My main argument against the AI doomsayers has so far been that the current scaling laws simply make runaway singularity style scenarios algorithmically impossible (if for each step of improvement you need 10x parameters and 100x training, you quickly run into a brick wall).

This is part of why I’m not worried about the current crop of generative AI. I am however both curious and concerned about what the tsunami of talent and $$$ chasing the current trend will achieve.

If this n^(4/3) alt transformer compute scaling is real (and there’s been many a pretender, so it’s too early to tell), then that could fundamentally change the overall AI scaling law, substantially lowering the brick wall.

And that could be a game changer.

My main argument against the AI doomsayers has so far been that the current scaling laws simply make runaway singularity style scenarios algorithmically impossible (if for each step of improvement you need 10x parameters and 100x training, you quickly run into a brick wall).

If this n^(4/3) alt transformer compute scaling is real (and there’s been many a pretender, so it’s too early to tell), then that could fundamentally change the overall AI scaling law, substantially lowering the brick wall.

I’m not worried about the current crop of generative AI. I am however both curious and concerned about what the tsunami of talent and $$$ chasing the current trend will achieve.

And I guess this may (or may not) be one of those game changing insights.

Jonathan Haidt has an interesting response to this (and related) criticisms: https://jonathanhaidt.substack.com/p/why-some-researchers-th...

Post a show hn?

(See the “show” tab in the top toolbar)

So not his son and no legal reporting requirement then?

The reporting is not incontrovertible.

It’s biased and heavily politically motivated.

Undermining SCOTUS because they don’t like the 6-3 majority (which I’m not a huge fan of either) is sacrificing the country for their short sighted political goals.

Is that the nephew he didn’t adopt? I.E. a child he was caring for, but not his actual son?

Since when is helping an orphan a cause for outrage?

Both matter, and returns fall off as you go further in one but not the other. The Chinchilla paper[0] established a simple scaling law for Large Language Models: model size and training tokens should grow at the same pace.

Compute is then proportional to the product of model size & data quantity.

That said, quality of data also matters a lot - OpenAI has had human labelers produce the data for their Reinforcement Learning from Human Feedback (RLHF), which has probably had a disproportionate impact on the success of ChatGPT compared to previous models, but that data is probably O(1%) of what they trained on.

At this point I'm guessing OpenAI are limited by both data & compute. Rumor has it they're training the "next big thing" now and it won't finish until December. If they had more compute they could presumably finish sooner, and if they had more data they would presumably let it train longer.

[0] https://arxiv.org/abs/2203.15556

From the project page: pronounced as "RwaKuv"

That is still quite challenging to pronounce, maybe one of "rwkv" -> "raw-kv" -> "rawk-v" -> "rock-v"?

Isn’t their point that humans also do that? Without it necessarily being called copyright infringement?

Lots of subtleties: fair use, what’s not fair use, etc?

It does seem like we’re in new territory, I can see merits on both sides and I guess we’ll figure it out via the courts / new laws over the next couple of years.

Escalation of commitment to a failing course of action. Has the money spent so far produced anything tangible & usable? Or is it all reports & studies?

I'd also really have preferred $, but apparently there's some backwards compatibility issue with using that character?

Note when this was written: 27th Jul 2020, i.e. ~2.5 years ago.

Pythons are snacking on GPS-wearing opossums that give up the python’s locations. It’s done to find the python, which is an invasive species they’re trying to control.

Death is short term hard, long term easy. Life is all over the place - sometimes hard, sometimes easy, usually misunderstood and massively influenced by whatever happens to be one’s current mood. Some have stable moods, some swing, some are more/less influenced by sleep, exercise & blood sugar level.

Ethics, morality, and philosophical approach in general typically evolve over a lifetime. Maybe your current one is incompatible. Ok. Maybe it’ll be the same forever. Ok.

But then again, maybe it won’t? Are you still young enough to be sure? (I’ve been so sure of so many things...)

Personally I found some semblance of peace when I finally accepted who and what I am, tried less to fit in and more to live & let live, and found a sleep, exercise & eating pattern that worked well.

Maybe you would too?

This sounds a lot like how my kids will listen to a teacher/coach, but not their parents...

?

It only resolves the links navigated to, right? It shows you the interstitial page, but only when you try to navigate to the link.

It does enable you to get the clean link and share that.

Being able to "refuel" quickly is a pretty big plus, so in the balance it could still be better to accept some moving parts in exchange for that.

And it's easy to imagine retrofitting a gas station to all of a sudden have charged electrolyte instead of gas, but I'm sure that has all sorts of additional complexities that I know nothing about... ;)

Does it really make sense to map each controller method to a lambda? Wouldn’t it make more sense to have gateway route to a single function per controller and then do the method routing inside that function? You’re going to get a lot of lambda functions otherwise.

Chargeback?

For clarity: the proof hinges on (√2 - 1) < 1 (so that k’ < k) and √2 being a square root so that you can multiply k’ by it and indeed get 2 k - k √2.

This is why it also works for √3, but doesn’t work for √5 or 1.5

Sounds like one of those cases where convenience trumps security for business users and the engineers keep their palms close to their faces.

As anyone who has ever faced a “rewrite from scratch or refactor”-decision knows (or eventually finds out), beyond some level of system scope & complexity a rewrite from scratch is not really feasible anymore. At least not without cataclysmic disruption and/or enormous delays.

America isn’t perfect. Nothing is. But boy the stuff we take for granted and forget all about.

Had exactly that bug in production, was using ruby on rails & active merchant, and some version change in ActiveMerchant switched from cents to dollars for one of the integrations.

Our test harness didn't catch it (weird combination of reasons, too long ago for me to remember the details) & it rolled out.

Shortly thereafter I get an anxious customer call that we'd charged their debit card $2500.00 instead of $25.00 and they'd gotten an overdraft notice. At first I was incredulous ("how is that even possible!?"), then I remembered that we'd just version bumped ActiveMerchant.

My endocrine response as I realized what must have happened was amazing to experience - the sinking feeling in my gut, hairs standing up, sweaty palms, dread, pupils dilating, and my internal video camera pulling back poltergeist-style in a brief out-of-body experience.

Fun times. Live and learn.

The description of how the transition to flight actually works was a little fuzzy - any video/animation out there?

Edit: https://youtu.be/4QaTwebs5nY - makes the mechanism much more clear.

> With enough traffic, their edge servers will keep connections open to origin.

Except AWS CloudFront when used with a custom origin server (whole site acceleration) - their "clever" load distribution algorithm requires inordinate amounts of traffic for effective origin connection re-use.

I run multiple sites in the top 100k websites globally and am getting effectively zero origin connection re-use. Badgered them about it but all I got was a wontfix :-/

Sigh.

The danger is less with automation and more with the speed of automation.

It's hard to retrain. When the change happens between generations there's a natural progression with parent: stable hand, child: mechanic.

When it happens 1-2-5 times in a person's career, then those are some really painful resets.

Sure, some will swim. But the social contract with democracy isn't that "some" swim, it's that "almost everyone swims" (which we generally call "everybody", but that's a separate discussion).

Long term and across society automation is absolutely positive sum. But short-to-long term for those on the short end of the stick it's clearly not.

The approach to edge-ifying SQLite taken in [1] looks quite promising - using FoundationDb as the storage handles a lot of hairiness.

1: https://github.com/losfair/mvsqlite