[munk-a]

Both numbers are entirely ludicrous - highly skilled people are certainly quite valuable. But it's insane that these companies aren't just training up more internally. The 50x developer is a pervasive myth in our industry and it's one that needs to be put to rest.

[__turbobrew__]

The ∞x engineer exists in my opinion. There are some things that can only be executed by a few people that no body else could execute. Like you could throw 10000 engineers at a problem and they might not be able to solve that problem, but a single other person could solve that problem.

I have known several people who have went to OAI and I would firmly say they are 10x engineers, but they are just doing general infra stuff that all large tech companies have to do, so I wouldn’t say they are solving problems that only they can solve and nobody else.

[remus]

I think you're right to an extent (it's probably fair to say e.g. Einstein and Euler advanced their fields in ways others at the time are unlikely to have done), but I think it's much easier to work out who these people are after the fact whereas if you're dishing out a monster package you're effectively betting that you've found someone who's going to have this massive impact before they've done it. Perhaps a gamble you're willing to take, but a pretty big gamble nonetheless.

[saagarjha]

OpenAI is taking bigger gambles already, though.

[scottyah]

It's apparent in other fields too. Reminds me of when Kanye wanted a song like "Sexy Back", so he made Stronger but it sounded "too muddy". He had a bunch of famous, great producers try to help but in the end caved and hired the producer of "Sexy Back". Kanye said it was fixed in five minutes.

Nobody wants to hear that one dev can be 50x better, but it's obvious that everyone has their own strengths and weaknesses and not every mind is replaceable.

[lovecg]

Do other professionals (lawyers, finance etc.) argue for reducing their own compensation with the same fervor that software engineers like to do? The market is great for us, let’s enjoy it while it lasts. The alternative is all those CEOs colluding and pushing the wages down, why is that any better?

[rester324]

Mmmmhm. You could have made this argument about 2 years ago, and it would have been credible. But you are making this argument now, when literally hundreds of thousands of engineers are let go in the last few years just in the US alone...? I am not sure how such an argument holds up in such circumstances...

[kridsdale1]

Talent and skill are a power-law, just as they are in basketball.

The United states has tens of millions of skilled and competent and qualified people who can play basketball. 1000 of them get paid to play professionally.

10 of them are paid 9 figures and are incredible enough to be household names to non-basketball fans.

[bitexploder]

The 50x distinguished engineer is real though. Companies and fortunes are won and lost on strategic decisions.

[kridsdale1]

Dave Cutler is a perfect example. Produced trillions of dollars in value with his code.

[a4isms]

> The 50x developer is a pervasive myth in our industry

Doesn't it depend upon how you measure the 50x? If hiring five name-brand AI researchers gets you a billion dollars in funding, they're probably each worth 1,000x what I'm worth to the business.

[belval]

> it's insane that these companies aren't just training up more internally

Adding headcount to a fast growing company *to lower wages* is a sure way to kill your culture, lower the overall quality bar and increase communication overheads significantly.

Yes they are paying a lot of their employees and the pool will grow, but adding bodies to a team that is running well in hopes that it will automatically lead to a bump in productivity is the part that is insane. It never works.

What will happen is a completely new team (team B) will be formed and given ownership of a component that was previously owned by team A under the guise of "we will just agree on interfaces". Team B will start doing their thing and meeting with Team A representative regularly but integration issues will still arise, except that instead of a tight core of 10-20 developers, you now have 40. They will add a ticketing to track change better, now issues in Team's B service, which could have been addressed in an hour by the right engineer on team A, will take 3 days to get resolved as ticket get triaged/prioritized. Lo and behold, Team C as now appeared and will be owning a sub-component of Team B. Now when Team A has issue with Team B's service, they cut a ticket, but the oncall on Team B investigates and finds that it's actually an issue with Team C's service, they cut their own ticket.

Suddenly every little issue takes days and weeks to get resolved because the original core of 10-20 developers is no longer empowered to just move fast. They eventually leave because they feel like their impact and influence has diminished (Team C's manager is very good at politics), Team A is hollowed out and you now have wall-to-wall mediocrity with 120 headcounts and nothing is ever anyone's fault.

I had a director that always repeated that communication between N people is inherently N² and thus hiring should always weight in that the candidate being "good" is not enough, they have to pull their weight and make up for the communication overhead that they add to the team.

[kridsdale1]

Have worked in BigCo three times scaling teams from 5 to 50 people. This post is bang on.

[hadlock]

You have to out-pay to keep your talent from walking out the door. California does not have non-competes. With the number of AI startups in SF you don't need to relocate or even change your bus route in most cases.

[darth_avocado]

This. The main reason OpenAI throws money at top level folks is because they can quickly replicate what they have at OpenAI elsewhere. Imagine you have a top level researcher who’s developed some techniques over multiple years that the competition doesn’t have. The same engineer can take them to another company and bring parity within months. And that’s on top of the progress slowing down within your company. I can’t steal IP, but but sure as hell can bring my head everywhere.

[epolanski]

This is also a good reminder of how there's no moat in AI.

I'm glad if US and Chinese investors will bleed trillions on AI, just to find out few of your seniors can leave and found their own company and are at your level minus some months of progress.

[xur17]

If it's an all out race between the different AI providers, then it's logical for OpenAI to hire employees that are pre-trained rather than training up more internally.

[epolanski]

50x devs are not a myth.

In any case the talent is very scarce in AI/ML, the one able to push through good ideas so prices are going to be high for years.

[rester324]

I think there is no evidence of any type of 50x devs. There is not even proof of 10x devs. So if there is no evidence, why is that not a myth?

[epolanski]

Of course there is.

There's always individuals, developers or not, whose impact is 50 times greater than the average.

And the impact is measured financially, meaning, how much money you make.

If I find a way to solve an issue in a warehouse sparing the company from having to hire 70 people (that's not a made up number but a real example I've seen), your impact is in the multiple millions, the guy being tasked with delivering tables from some backoffice in the same company is obviously returning fractions of the same productivity.

Salvatore Sanfilippo, the author of Redis, alone, built a database that killed companies with hundreds of (brilliant) engineers.

Approaching the problems differently allowed him to scale to levels that huge teams could not, and the impact on $ was enormous.

Not only that but you can have negative x engineers. Those that create plenty of work, gaslighting and creating issues and slowing entire teams and organizations.

If you don't believe in NX developers or individuals that's a you problem, they exist in sports or any other field where single individuals can have impact hundreds of thousands or millions of times more positive than the average one.

[rester324]

I asked if you can prove there are 10x or 50x programmers. You shared anecdotes and theories. I will rather wait until you share some evidence.

[epolanski]

If you don't see the evidence of different individuals having very different productivity in every field, including software, (measured in $/hr like every economist does btw) that's a you problem.

Of course different scientists with different backgrounds, professionalism, communication and leadership skills are going to have magnitude of orders different outputs and impacts in AI companies.

If you put me and Carmack in a game development team you can rest assured that he's going to have a 50/100x impact over me, not sure why would I even question it.

Not only his output will be vastly superior than mine, but his design choices, leadership and experience will save and compound infinite amounts of money and time. That's beyond obvious.

[SiempreViernes]

That devs might show a 10x spread in time to completion on some task (the mythical man month study) is quite a lesser thing than claiming the spread comes from something inherent to the devs that got tested.

As for your various anecdotes later, I offer the counter observation that nobody is going around talking about 50x lottery winners, despite the lifetime earnings on lotteries also showing very wide spread:. Clearly observing a big spread in outcome is insufficient evidence for concluding the spread is due to factors inherent to the participants.

[causalmodels]

These numbers aren't that crazy when contextualized with the capex spend. One hundred million is nothing compared to a six hundred billion dollar data center buildout.

Besides, people are actively being trained up. Some labs are just extending offers to people who score very highly on their conscription IQ tests.

[saagarjha]

I think the unfortunate reality is that training someone to reach the frontier is time taken away from actually pushing it. The opportunity cost alone is worth millions to them.

[charcircuit]

It's not a myth and with how much productivity AI tools can give others, there can be an order of magnitude difference than outside of AI.

[xnx]

> training up more internally

Why would employees stay after getting trained if they have a better offer?

[munk-a]

They won't always. You'll always have turn-over - but if it's a major problem for your company it's clearly something you need to work out internally. People, generally, hate switching jobs, especially in an uncertain political climate, especially when expenses are going up - there is a lot of momentum to just stay where you are.

You may lose a few employees to poaching, sure - but the math on the relative cost to hire someone for 100m vs. training a bunch employees and losing a portion of those is pretty strongly in your favor.

[coolspot]

A tamper-proof electronic collar with some C4.

[gmerc]

Zuck decided it's cheaper than building another Llama