Swizec

I feel like that wasn’t there when I started writing my comment. I also have a bad habit of quickly posting and then adding over a few minutes.

Glad we agree :)

> As far as I could tell, nothing was substantially different from the Ellsworth translation and the Fable translation.

Crucially the full translation was part of ChatGPT’s training set. Recall is a pretty solved problem in machine learning.

How well does it translate a French novel published yesterday? Where neither the original novel nor any translations are in the training set yet? Or might not even exist!

I tried asking ChatGPT to translate a letter I wrote in Slovenian this weekend. It got the general gist but missed a lot of the nuance. Completely missed several of the little touches of tone where the right choice of synonym conveys a whole bunch of information.

> Until that time, such countries would just not have access to systems before they were bulletproof.

Correct, most jurisdictions do not allow businesses which cannot be held liable for their actions. This is pretty core to a modern society.

Imagine if a company selling Knicks tickets was not expected to then actually provide said tickets and there was simply nothing you could do about it. Oopsies our sales page is for entertainment purposes only

To be fair, the internet has spent some 30 years figuring out how this works and it’s still not fully resolved. For the most part we’ve agreed that companies must follow the laws of both where they live and where they operate. This wasn’t always obvious!

> at the scale that google is at, it doesn't make an ounce of sense to hold a company liable for a potential mistruth

If a Google employee (like a support agent) says a mistruth, the company is liable and you can sue. They can’t just say “hihi oopsies our support agents are useless”

Good. The true mark of AGI is when a company accepts liability and doesn’t bury “for entertainment purposes only” deep in their TOS. Same as it works with employees.

Same for self-driving. Your car is not self-driving until it accepts liability and you count as just a passenger.

But watch as Germany soon loses AI Google results.

> Did they think, as they worked to transfer final say from users to corporations, by technical means, that politicians couldn't transfer that control to themselves by political means?

Makes me think of the most sobering line I ever saw in a museum (Berlin): The biggest atrocities were committed by people with a spreadsheet and a performance goal.

> But then why didn't they cut it into 10 pieces - https://etc.usf.edu/clipart/40600/40610/pie_01-10a_40610.htm ?

Ever tried cutting a cake? It’s a lot easier to visually judge half of a circle segment. You’d need a compass to get accurate tenths (or fifths) and I imagine it is generally frowned upon if some tenths are a lot smaller than others (happens a lot with cake)

> Now, a comparable energy budget we used to use "deciding what to build" (because labour was scarce) is now energy we can divert toward "unbuilding stuff that was a mistake"

The hard part is that you very quickly become Salesforce or Jira or <insert large confusing product>.

You have thousands of users who love your product and pay lots of money and find the features absolutely essential to their workflow. Everyone says your product is bloated and has too many useless features, if only you could delete a bunch of crap they don’t use your software would be perfect.

They all use a different 20%. Delete a feature, lose a fifth of your users.

> Ever considered fan noise while shopping for laptops? If so, then you don't want to live within a mile of a datacenter.

Or we could regulate that data centers be sound proof.

These are things we can solve. They just cost a little money so businesses will fight tooth and nail against it. But hey look we also used to dump slaughterhouse refuse and factory runoff straight into the river in the middle of cities. We don’t do that anymore because at some point it became illegal.

Easy peasy. Just make the things you don’t want businesses to do illegal and they’ll stop doing them.

We could even regulate that all data centers have a large public park and green space on its roof! Or be covered in solar panels to make its own power. Or a huge parking lot. Whatever we need or wish for, the billion+ dollar investment into the data center can provide.

I was doing this in 2003 and my computer was also the internet/network router for our house. When that thing was down, you had no access to external information that you didn’t pre-save somewhere.

One time I forgot to install network drivers and had to download them through my flip phone via GPRS and then awkwardly load onto the computer via a clunky USB connection. Fun times.

Also my English wasn’t this good yet. I’m sure it would’ve been a lot easier had I actually understood all the tutorials and documentation fully.

> The other 99% who were into yoyo-ing back then are now into TikTok, that's all.

Hey dude, some of us were yo-yoing while waiting for Gentoo to build from stage 0. Compiling an OS on a single-core Athlon takes time.

For the 3 days it takes to build all the way up to KDE, you have no computer. Hope you didn’t forget something

> Do you still have problems with doomscrolling addiction?

The new twitter algorithm fixed that for me. They made the feed so fundamentally uninteresting and full of slop* that the issue resolved itself.

I now doomscroll slack and discord

* by slop I mean content written for the sake of being content.

> You can use it to make the entire phone greyscale, which entirely defeats a huge range of techniques that apps and feeds use to draw your attention

As a counter-anecdata: I've had my phone in greyscale for a few years now. At first it worked amazingly, made me hate my phone and pickups dropped significantly. But over time I realized "Oh wait, 90% of my phone use is text and this is actually super nice for reading".

Now I use my phone just as much as before, except in greyscale.

> Until AI is basically in a stable, predictable, state of improvement or stagnation, these takes will continue to be pointless and most likely completely wrong.

Little thing to keep in mind about AI: a technology is only called AI while it doesn’t work yet. Once it works reliable, we give it a proper name and something else becomes AI.

> 'build a lot of homes at once'. That creates the suburbs which

What a silly American way of thinking. Build a lot of homes at once creates high rise neighborhoods. We've had these in Europe since the 60's, they are great. Asia has taken it to the extreme in recent decades.

A couple high rises give you a few hundred residential units in a completely walkable neighborhood.

Here's an example from Ljubljana, built between 1977 and 1987. Houses 18,000+ people on 150 acres. https://sl.wikipedia.org/wiki/Nove_Fu%C5%BEine

Here's a more sciency source. Over 3000 residential high rises were been built in Europe in the 2010's

https://www.alexandrinepress.co.uk/built-environment/high-ri...

Another fun example: The tallest skyscraper in Europe will have 260 apartments and 107,000+ sqft of communal space https://www.domusweb.it/en/news/gallery/2025/09/26/benidorm-...

Oh I’m on essentially permanent 2nd tier oncall and occasionally 1st tier. Pagerduty has an exception configured. If you message me on slack it shows up on my home screen and I will probably notice in the next 15 minutes or so.

I find polling less disruptive because my phone or watch are almost always nearby. Being in control of when to get interrupted feels better than having stuff constantly pop up while you’re doing something.

Just holding my partner’s phone spikes my cortizol. You try to type 2 sentences and get 5 notifications for random other apps and messages popping up on screen in the meantime. Literally one notification every 10 seconds on average. It’s horrible

> The point of notifications is the convenience of not having to constantly check your phone for every single app you have (amazon delivery? just eats delivery? uber booking? claude finished its task?).

My phone has been on DoNoDisturb since 2010 or so. Here's the reality: I don't check for any of those things. Delivery drivers can ring the door bell. If I'm very hungry I'll keep the app open and check where they are. I literally do not care to be notified about any of the things that apps want to notify me off.

Anyone who cares to reach me knows to ring the phone twice in case of emergency to get through DnD. Anyone else? The best time to call is text me. Or schedule a time.

As for Claude, the point of clankers is that they work in the background. The robot can wait, their infinite patience is a feature.

> An actually smart entity would not do that if not acting maliciously.

We pay per token and every entity falls to the level of its incentives.

I’m eastern-ish european, is it even racist to say that tech talent in the region is through the roof but for various accidents of history, the best opportunities available to talented people are in cybercrime (both sides)?

Not everyone has a hundred tech unicorns in their back yard. I think my country (Slovenia) produced one in its entire history so far and even that was mostly in the US

It’s really hard to get people excited about not having jobs when you design a whole society around the idea of having a job and make life exceedingly miserable for anyone who doesn’t

You can’t push both “If you dont work, you dont eat” and “Nobody needs to work anymore” propaganda at the same time. Gotta choose

> So long as the code works and meets all stated requirements and is secure and performant and uses good abstractions and is not full of hacks, then it's ok to let go

That is the problem. The code often is full of hacks and bad abstractions. LLMs write code like a junior or mid-level engineer – perfectly overfitted to today’s request. Oh you need to work on this code tomorrow and there’s a laundry list of future requirements? Throw away and rewrite, I guess.

You can most easily see this when you ask LLMs to write tests. They have a tendency to write convoluted tests that absolutely definitely pass. Even when you know the code has a bug, they’ll write the test in a way that fits the code as written and passes. Because they know tests should pass.

Getting an LLM to write a failing test against a currently working function because you know the business requirements have changed is like pulling teeth.

You don’t see writing about this stuff because it doesn’t neatly fit in an article or video (I’ve tried). Plus it goes against the zeitgeist so you’d never get traction (even if people write these posts, we don’t see them)

> How are we really supposed to grasp their actual capabilities when no one will actually cite specifically what mistakes they are making.

The mistakes they make are pretty subtle. Coding with LLMs can be like that scene in Whiplash – <excellent drumming >, not quite my tempo, <excellent drumming >, downbeat on 18, <excellent drumming>, you’re rushing, <excellent drumming>, dragging, …

Like yeah it produces working code almost always and the code usually does what you asked. And yet it makes you want to throw a chair because it’s not quite right in frustrating ways and it doesn’t even have the taste to know how it’s wrong.

> Huh... I'd always heard that a golf ball's dimples help reduce drag?

Yep also vortex generators in cars have become common. So common that they've filtered down to after market parts you can put on a honda civic

Vortexes break up large air pockets and reduce drag.

You can marry in the US and have kids and still move somewhere else 20 years later. Don’t Americans move to Florida or whatever to retire? If you’re moving that far you can just as well leave the country lol

> 300k in SF or NYC is FAR from early retirement unless you live 'frugally' - Manhattan average rents are 5K for 1 bed

You don’t have to retire in the US. As others have pointed out, nobody comes here for the lifestyle.

Immigrants like us are literally the holy grail of immigration. Come in during our most productive years, work hard for 10 to 20 years, go back home before you need any of the social and health care stuff you paid into.

> 401(k) and IRA’s were not a gift

No not a gift but they sure do align incentives

> will at some point in their life belong to that group which owns 90% of the means of production

Would love to but I'm afraid you grossly misunderstand how the power law distribution works. Unfortunately the billionaires are running away from us exponentially faster than we are catching up.

Speaking as someone who will almost certainly become a multi millionaire in the next decade or two. (even if I stop adding more savings manually)

But crucially this only happens if there _isnt_ a big revolution in USA. If there is, I'm fucked.

> Meanwhile tech workers in the US spend all day online defending billionaires

Tech pulled a great trick here: equity.

America as a whole pulled a similar trick: 401k

It’s hard to fight the billionaires when all your money depends on not fighting them.

> That implies Karpathy is either dumb or desperate and he is neither of those by a long shot.

No it implies that he is more valuable for being famous than the hands-on work he can produce. This is the IC endgame

> And to me, AI should best be used to add rocket fuel to existing practices

The more your codebase follows best practices and consistent patterns, the better AI will do and the faster you can move.

Same as humans really, just even faster. I'm also excited that people are finally writing docs and without even any flogging! They're calling the docs "skills" but hey whatever works