input_sh

Say you use a magic wand and build 15 new state-of-the-art factories tomorrow. Who's gonna run them? Does any location in the US have enough qualified workers that can simply take over and produce RAM in them from day 1 with no major fuckups?

No, you need a ton of time to teach thousands of people how to run those 15 factories. To even begin to teach people, you need to have 1 factory up and running. That 1 factory is at first going to be run by some of their existent workforce that they temporarily migrate from South Asia. Only then can they start to teach local populace how to run those factories on their own.

This is why it's much cheaper to simply build an additional one in South Asia than it is to build more than one in a whole new location. South Asia already has a bunch of workers that know what they're doing because they've been doing it for a long time. Build a new factory, promote some of your existent workforce up the chain, fill the lowest positions with fresh graduates that are gonna be equally good every year and you're good to go. It's nowhere near that simple in a brand new location, where even the most optimistic scenario would take longer than a decade to produce a meaningful amount of output.

Building a factory is one thing, they can have 50 of them built, but that doesn't mean much if all 50 together amount to like 0.1% of the company's output.

Once those factories scale up to 1-2%, then we can start considering that they've actually built a domestic supply, but that's a whole different goal than simply building the factories. Building factories is trivial. Making them output something is also "trivial". Scaling that up to a meaningful amount is a whole different, much harder goal to accomplish.

https://www.micron.com/us-expansion/id

> Micron has already achieved key construction milestones on its first Idaho fab with DRAM output scheduled to begin in 2027.

https://investors.micron.com/news-releases/news-release-deta...

> Production is expected to start in 2030 with the fabs ramping throughout the decade.

Until they start outputting DRAM in any meaningful quantity, they're not relevant.

Because every other language has far worse moderation, and you can pretty much guess how good the moderation is simply by asking yourself how relevant that version of Wikipedia even is in the first place?

I can understand 6 different versions of Wikipedia and my experience is the complete opposite of what you're insinuating, English version beats the other five 99.8% of the time even the topic at hand is completely local.

That said, I am biased as I maintained quite a few of them years ago and am happy to see today's youngling maintain this tradition of low-effort contributions to make a source that's better than a search engine when looking for stuff in a specific niche.

https://www.goodreads.com/quotes/7870768-never-believe-that-...

On top of every version of Gemini, you also get both Claude models and GPT-OSS 120B. If you're doing webdev, it'll even launch a (self-contained) Chrome to "see" the result of its changes.

I haven't played around Codex, but it blows Claude Code's finicky terminal interface out of the water in my experience.

I'm gonna single out Grokipedia as something deterministic enough to be able to easily prove it. I can easily point to sentences there (some about broad-ish topics) that are straight up Markov chain quality versions of sentences I've written. I can make it say anything I want to say or I can waste my time trying to fight their traffic "from Singapore" (Grok is the only "mainstream" LLM that refuses to identify itself via a user agent). Not really a tough choice if you ask me.

Instead, many, many websites (especially in the music industry) have some sort of funky API that you can only get access to if you have enough online clout. Very few are transparent about what "enough clout" even means or how much it'd cost you, and there's like an entire industry of third-party API resellers that cost like 10x more than if you went straight to the source. But you can't, because you first have to fulfill some arbitrary criteria that you can't even know about ahead of time.

It's all very frustrating to deal with.

There's simply no way to package that which doesn't make the other side think that I'm gonna steal company's time at best and that I'm only looking for like a temporary gig until it takes off at worst.

Usually you only get some specially-crafted demo files that are capable of fooling you.

LLM didn't discover this issue, developers found it. Instead of fixing it themselves, they intentionally turned the problem into an issue, left it open for a new human contributor to pick up, and tagged it as such.

If everything was about efficiency, the issue wouldn't have been open to begin with, as writing it (https://github.com/matplotlib/matplotlib/issues/31130) and fending off LLM attempts at fixing them absolutely took more effort than if they were to fix it themselves (https://github.com/matplotlib/matplotlib/pull/31132/changes).

I don't, and since neither of us can know for sure given the info we've been given, it's useless for us to argue about our opinions.

They had plans to bomb something south of that airport, they had to postpone those plans now that the info is public enough that whatever their target was is definitely aware of those plans.

You need to look at both sides of the transaction to figure out the total.

You know, the same country whose former prime minister is this person: https://jmail.world/person/ehud-barak

There's still great stuff out there, it's still as hard to find and navigate as it has always been. It's still as shady and as illegal (if you care about copyright) as it has always been. Most people still don't bother to do it, you just became a part of that category.

Here, let me try snapping you out of that bubble a little bit and make you one of the today's lucky ten thousand: find a category that interests you on fmhy(.)net (SFW, I promise), see how long it takes you to spot something you had no idea existed outside your bubble.

There are plenty of them being built, yes. Some of them will even start outputting products soon enough. None of them are gonna start outputting products at a scale large enough to matter any time soon. Certainly not before 2030, and a lot of things can change until then which might make the companies abandon their efforts all together or downscale their investments to the point where that due date gets pushed back much further.

That's not even discussing how easier it is for an already-established player to scale up their supply versus a brand-new competitor to go from zero to one.

For example a lot of pro-OpenAI astroturfing really wanted you to know that 5.3 scored better than opus on terminal-bench 2.0 this week, and a lot of Anthropic astroturfing likes to claim that all your issues with it will simply go away as soon as you switch to a $200/month plan (like you can't try Opus in the cheaper one and realise it's definitely not 10x better).

Sure, the tiny rushes of dopamine are free, but that's just enough to get your foot into the door and pretty soon they won't hit the same. If you don't quit while you're ahead, it's either gonna financially ruin you or turn you into a zombie that never uses its brain for an extended period of time and you end up writing posts like this one to self-rationalise your completely irrational behaviour because you've been scared into thinking your career is gonna be ruined if you don't keep diving head-first into this addiction.

Not a coincidence that this is happening at the same time as "bet on anything" apps.

Once the website is designated as such, you're looked at with more scrutiny, have to comply to higher standards, and the exact remediation steps are decided on a case-by-case basis. All of the cases are chugging along, but not all of them are on the same stage.

If your website is not popular enough to be designated as VLOP, this law basically doesn't exist. It's not like GDPR in a sense that it defines some things everyone has to follow, regardless of your audience size.

Now let's assume you want to send a CD worth of data to Opus 4.6. 700 megabytes * $10 (price per million input tokens) / 4 (rounding down one megabyte to roughly 250k "tokens") = $1750. For Opus 4.6 to return a CD amount of data back to you: $37.50 * 700 / 4 = ~$6.5k.

A terabyte worth of data with a 50:50 input/output ratio would cost you $5.7 million. A terabyte worth of data with a 50:50 input/output ratio on gpt-5.2-pro would cost you $25.2 million. (Note: OpenAI's API pricing still hasn't been updated to reflect 5.3 prices.)

So we get layers upon layers upon layers upon layers upon layers of obfuscation to hide those numbers from you when you simply subscribe for a fixed monthly fee!