Live data from Hacker News

John Carmack Leaves Meta

facebook.com

681–690 of 1001 posts

Re: John Carmack Leaves Meta

#681

I’m a strong believer in dictatorship at companies. Every team has someone with ultimate decision making power, as does every project with a project leader, and above teams are leaders who have authority above them, and so on. Of course there are meetings and discussions but there is always someone with the authority to make a decision that everyone has to respect. Honestly it benefits everyone. ICs like that authori…

The trade off is decision cycle time. In a dictatorship, the dictator (and getting their attention) becomes a bottleneck.

Re: John Carmack Leaves Meta

#682
post #635
post #626

I see a lot of negative comment on the other thread and on twitter so I feel the need to comment as well. It’d be great to get the perspective from someone who worked there, but I can offer mine as an outsider (worked on a different team at fb). Back then, I would religiously read every workplace posts from Carmack (that wasn’t hidden to other orgs). I loved everything he was saying, I always found it really insightf…

> You’re supposed to lead without authority John seems to be particularly calling out that he was not able to influence the people with high enough rank to make decisions. So it sounds like he at least thought there was an authority structure there wasn't as flat as you describe.

There is always an authority structure in these kinds of companies. I'm personally familiar with Google, but this description of Facebook makes it sound similar...

The upside is that people, in theory, don't need to wait to be promoted to (attempt to) have an outsized influence on what actually happens on their team, in their org, etc. The downside is that the "real" power/influence structure can be a more nebulous/social game, rather than something concrete that is diagrammed out in a visible org structure. It was often said at Google that you get promoted to level X after already doing work at that level, which implies that the people with influence are running around influencing/leading before they've been anointed with an official title that "gives" them authority.

Reading Carmack's letter, it sounds like he consciously chose to prioritize his personal programming/engineering work over his social/influence/leadership at Facebook. In all my observations, past a certain level, you kinda have to choose one or the other. Both are valid choices.

Re: John Carmack Leaves Meta

#683

I think Carmack is a great engineer. Part of that is being great at communication. And he is! Usually. But this paragraph is quite hard to follow: > If I am trying to sway others, I would say that an org that has only known inefficiency is ill prepared for the inevitable competition and/or belt tightening, but really, it is the more personal pain of seeing a 5% GPU utilization number in production. I am offended by i…

Yeah this was really odd to read?

5% gpu utilization on your battery hungry mobile device? That's not bad it's good/bad??

Re: John Carmack Leaves Meta

#685

I’m a strong believer in dictatorship at companies. Every team has someone with ultimate decision making power, as does every project with a project leader, and above teams are leaders who have authority above them, and so on. Of course there are meetings and discussions but there is always someone with the authority to make a decision that everyone has to respect. Honestly it benefits everyone. ICs like that authori…

Finding good devs is hard enough without a dictatorial style of leadership. I find that people like democracy, for some reason.

Good leadership feels like a democracy while actually being a dictatorship.

Walk people towards the decision you want using a Socratic method and then let them take the credit for the decision making.

Of course sometimes you will get a situation where someone doesn’t arrive at the solution you want anyway so you need to excise a bit of hard power, but hopefully that is rare enough that people respect it when you do it.

Re: John Carmack Leaves Meta

#686

"I have never been able to kill stupid things before they cause damage" One of the most important things an organization can have is a "no man". A project that I was involved in when I worked at Google was completed on time primarily because we had a very senior (DE level) engineer tangentially involved in the project. He was near the end of his career, and he just didn't give a f* about politics. He'd sit in design…

The part about not giving a fuck is crucial. You don't get rewarded for preventing bad things.

In fact, you might feel like a hero for needing to put in OT to put out a garbage fire! Never mind that it shouldn't have happened in the first place...

Re: John Carmack Leaves Meta

#687
The real problem here is the memtaverse. This is a solution to a problem no one has and it's costing a vast amount of money for, well, nothing really.

Why does it exist? Simple. Meta needs a new monopoly since FB usage is declining and IG is getting eaten alive by Tiktok. Messaging (ie FB Messenger, IG Direct, WhatsApp) is not enough.

Meta has long seen VR then AR as the natural evolution from text -> image -> video. I think it's clear that the VR part of this at least is in error. Personaly I see VR as never being anything more than a niche. Somehow spending $20B+ a year on that without any kind of product-market fit or a vision for what problem this will solve for people is the problem.

AR is way more likely to have a future but the tech isn't there yet and there are doubts it'll be anything more than a niche either. This is a deep topic but projecting things onto real vision isn't exactly simple. Even something as simple as the color black is a problem. Focus is another giant problem.

For anyone surprised how a project can spend billions without producing anything, this is classic big company poorly defined project type stuff. A project will expand to fill available resources. Writing a blank check just increases the head count. It doesn't produce more just because you have more head count. People without clear direction will invent fake work for themselves. They'll solve non-problems, creates frameworks, add processes and so on as necessary.

Disclaimer: ex-Facebooker.

Re: John Carmack Leaves Meta

#688
post #635
post #626

I see a lot of negative comment on the other thread and on twitter so I feel the need to comment as well. It’d be great to get the perspective from someone who worked there, but I can offer mine as an outsider (worked on a different team at fb). Back then, I would religiously read every workplace posts from Carmack (that wasn’t hidden to other orgs). I loved everything he was saying, I always found it really insightf…

> You’re supposed to lead without authority John seems to be particularly calling out that he was not able to influence the people with high enough rank to make decisions. So it sounds like he at least thought there was an authority structure there wasn't as flat as you describe.

> authority structure there wasn't as flat as you describe.

It never is. In the absence of formal authority, people will unconsciously build an ad hoc structure on their own.

Re: John Carmack Leaves Meta

#689

Can someone explain me the 5% GPU utilization issue

For the monster hardware it has, Oculus (at least in games) can only rival a PS2 in terms of polygon count. Pair that with the 5% GPU utilization metric and you get that this device can handle a whole lot more load and it's just being used inefficiently. I'd even use the word 'wrong', people within Meta made 'good enough' libraries and called it a day, and he's mad about that.

This thing has billions of dollars poured into it, and all you get is PS2/PSP level graphics? No way, man.

Carmack of all people knows the kind of feats this machine is capable of, having worked complex optimization problems in the RAGE Engine and the old Quake engine too.

Re: John Carmack Leaves Meta

#690
The thing about 5% gpu utilization makes me wonder if all this talk about training LLMs on multi A100 clusters is largely unecessary and that it might be possible for everyday GPUs to reach say A100 levels simply by going from 5% utilization to 90%.
Post reply on HN