Zos_Kia

Microsoft avait expérimenté avec Data Center immergé

Pour le coup je pense que l’océan c’est le pire endroit possible pour dumper autant de chaleur. C’est vraiment très problématique surtout que ça nuit à la capacité de l’eau à absorber le CO2 de l’air donc ça a un effet composé.

Après pour les résaux de chaleur c’est super intéressant je ne sais pas dans quelle mesure c’est possible mais du coup un DC avec énergie décarbonée + production de chaleur pour la communauté locale ça devient assez attrayant comme proposition.

décentraliser les couts de l’IA ne la rend pas plus efficace. Le local c’est pour des questions de vie privée, pas d’environnement

On est quand même en droit de penser qu’il vaut mieux avoir des GPU à 400W qui font des spikes occasionnels et déversent quelques calories dans le buffer thermique d’une résidence (où elles sont compensées mécaniquement par moins de chauffage etc…), que des data centers en spike permanent qui dumpent des MW de chaleur dans leur environnement immédiat.

It was a mess from superfluous complexity from adopting every buzzword along his career, cloud, microservices, configuration management

That was the bane of my existence before AI and i suspect AI will only compound this issue.

If you do “artisan vibe coding”, acting like a very hands on CTO that challenges decisions and reviews most of the code produced, you get a modest productivity boost in the 20 to 40% range, and a large reduction in cognitive load which can help you think bigger thoughts on the longer term. The quality can be as high as you want it to be in that setup.

But if you do fully agentic unsupervised vibe-coding, it’s easy to get into a mess because it’s like having a team of junior/mids paid by the line churning out complexity all day long. The productivity boost can be a large multiple but the quality suffers because you have to ignore a lot of the stuff and the devil is in the details so he will certainly get you at some point.

I mean obviously mileage does vary from project to project and task to task, but i think you might be overestimating mid-level developers. Or you’ve been really lucky with your recruitment ! Cause i would describe them just the way you described Opus. Pretty eager, kind of try-hard, decent engineering chops but often misdirected with dumb approaches.

Of course my experience is limited and i’ve never really been in a managing role but i’ve been the adult in a fair number of rooms and i’ve done my share of “grooming sprints” and dispatching tasks.

That being said, there are projects that are horribly reluctant to agentic coding. It’s pretty rare as most codebases nowadays are bog standard and rely on roughly the same abstractions, but i’ve seen it happen. It can come from the complexity of the domain, or of the codebase, or from the way documentation and tribal knowledge clash, or a myriad other reasons. Often it’s the kind of projects that require more mature devs and can’t really onboard juniors/mids.

When it digs itself into a hole, it’s very bad at trying to amend the mess that has accumulated

Oh yeah definitely. Once it’s in the hole you better scratch that branch off and restart with more specific instructions cause agents are very “additive”, they don’t often think to remove stuff and change their approach. Again, kind of like mid devs once they’re committed to an implementation plan.

Aha thanks for sharing that’s a cool anecdote. But i think my point still stands, as there are thresholds effects in LLM “intelligence” which don’t directly map to the RAM comparison.

Opus 4.6 is comparable to a mid-level developer. It requires some guidance and will sometimes get things wrong, but is also suitable to work in most business environments: most projects are not that complicated or high stakes in the first place.

In the future you’ll probably have Opus 7.5 or some shit, which will be at a mega-senior level but also considerably more expensive. And given the price difference, companies will suddenly discover that they don’t really need expert level coding at a high price tag, and that a reliable workhorse at a fraction of the cost is largely enough for their needs.

Not sure what you’re referring to?

I think the issue is also that you need some serious hardware to get good inference speed when your devs are working, but then most of the time this hardware will be under utilized.

That being said you can get good performance from indie inference farms, at a fraction of the cost of the big US labs. I think it’s a great compromise and in a few months the open models will be near parity with opus 4.6 which is really all you need for most tasks.

It’s kind of standard practice at this level. They get a million resumes a week in inbound, of which they hire very little, and they have an army of recruiters who go searching for people in the industry, who for the most part already have a job.

It’s way more profitable to let small startups grow juniors into competent employees and pluck them when they’re ripe, than to take risks on junior profiles. (not that i condone it but it’s a sad truth of recruitment)

Exactly. Mega corps don’t touch entry level work, they do exactly as you say and pay a friend of theirs so that this friend can steal wages from poor people and guarantee them floor-level prices.

Honestly it doesn’t matter if you make 300k a year at Meta or $12 an hour at Walmart

It absolutely does. If you can make 300K at Meta you can make 250K someplace else. Missing out on those extra 50K is nowhere near an existential threat to you.

If you make 12$/h at Walmart i am not blaming you for taking what you can and making a living of it. If you work for a Meta subcontractor in Kenya doing 300$ a month i am not blaming you.

But if you choose to work for an objectively evil company while you have other options (which you definitely have if EvilCorp is even considering your resume), and you make that choice just to maximize your discretionary spending… then you can’t play the solidarity card.

Are you under the impression that only Amazon, Microsoft & Walmart are hiring ? Did every software company suddenly vanish while i was napping ? Sorry but there are a million companies hiring the same kind of profiles as Meta, which are not evil monopolistic mega-corporations.

Also what’s Walmart got to do with this ? You think greeters and cashiers are hired from the same pool as Meta product owners ?

I’ve known people with experience and degrees that have been searching for like 6 months

If that is the case then Meta probably won’t touch their resume with a 10-ft pole.

I mean if I worked for a soul sucking corporation (pick one Amazon, Microsoft, Apple, etc.) and another shitty corporation (Meta) offered me a substantial pay raise why wouldn’t I take it?

But that’s not the real scenario. Monstruous companies aren’t the only jobs around. There’s a bajillion smaller and more ethical companies that hire the same kind of profiles as Meta.

Also the mega-corporations tend to not poach small fish from each other. Sure they’ll go after each other’s top performers but there’s a tacit agreement that “normal workers” are not fair game.

meta doesn’t only have high level jobs

The majority of their hires are experienced profiles. “Bog standard” worker in Meta is already a pretty high paying job as entry level work is massively outsourced.

all who work there are some kind of imaginary elite?

That’s not at all what i’m saying. I’m saying if you can land a job at Meta then you can absolutely find a job elsewhere. There is nobody in the world who is stuck between an empty belly and accepting a position at Meta. They’re just stuck between making good money somewhere, and making better money at Meta.

That’s bullshit. Almost all meta hires are poached, and they’re not coming for you if you’re not in a position to make good money.

The vast majority of these people chose to leave an already cushy job to get in there.

I love that their moat has been bridged by open models as soon as it was dug. At the current rate, you just have to wait six or nine months and open models will be at opus 4.6 level which is really all you need for most applications. After this I don’t see how the big labs could ever recoup their losses.

That’s not really what’s happening they’ve been trimming a lot of middle manager fat too

This reminds me of Barney Stinson

You don’t hire interns for productivity

Because it’s unethical. I’ve been in business for 10+ years but i never hired an intern because i don’t find it fair to make someone work for less than minimum wage, and i don’t have the structure required to really teach them anything. I have bad fundamentals and only ever learnt by doing, so having an intern while it may help me wouldn’t really help them and that’s not a deal i’m willing to make. Probably why i’m not super successful lol

That being said, i don’t see any problem with making a GPU cry somewhere in California for my menial tasks. And it’s tremendously effective too, for a hundred bucks a month i get a lot of shit done that would take me ages. I don’t give it access to anything critical so it can’t fuck my shit up and i come out on top as long as the tokens are subsidized by dumb VC money.

Yup it’s one of those examples where the mystery is way better than the contrived answer.