Comment by biztos
3 years ago
I'm sympathetic to this argument, because I know both low-quality and high-quality engineers at most FAANGs, but in principle isn't offering a lot of money a good way to attract talent?
In the case of OpenAI you also have interesting tech and a brand that will massively accelerate your career if you want to stay in that field. So while yeah, you have to hire the best people; and OpenAI like everyone else will be paying $LOTS to a few useless engineers in the mix, I think "$900K and everyone knows it" is a pretty good substitute for talent-spotting, which anyway can't be bought.
I worked with the enterprise sales team for awhile when I worked at Udacity and we did some research on job motivations to help answer questions about recruiting & retaining good engineers. We found that salary needs to be “enough” (varies by local market) but that people will literally relocate their families and lives to work on problems that they find interesting and that are a unique opportunity. The salary bump needs to be very large to achieve the same effect as “amazing opportunity”.
Both may be benefiting openAI here. There’s lots of places to work on LLMs but “GPT” is a product/brand that people have heard of, and if OP is to be believed then they certainly seem to be paying “enough”.
Sort of, you will probably attract the best talent by having the hardest problems, fostering a culture of bottom-up innovation, and giving your engineers a lot of freedom rather than micro-managing them. The pay is closely related, but you also get a lot of people trying to game the system and optimize for max pay, without necessarily the skill or creativity you're looking for. Said another way: top pay attracts people looking for top pay, for better or worse. There's probably a strong correlation between "best" people and high pay, but it's also hard to quantify "best".
How do you quantify best? Number of degrees? Publications? Association with prestigious institutions? Past work experience at top companies? Speed of problem solving? All of this is gameable once it starts being _measured_ and enough incentives exist for people to devote their life to winning the "game".
However, if you happen to hire a math olympiad winning PhD with numerous publications from a tier-1 research institution with a known track record in industry, it would be hard to argue they aren't the best. But success breeds success, and top people will keep being poached to other top places. Kinda how money makes more money.
> How do you qualify the best?
We’ll, by the number of Leetcode Hards they’ve completed, of course! /s
How do you quantify best?
The bests have cool stories to tell.
Right, but you have to be in a position where you get to do the work in the first place, which requires passing some reputational and skill bar. Or, you could do something which is cool to you personally and someone might not share that feeling and be unimpressed.
As a concrete example, I've gotten more accolades for silly personal projects that sound impressive, like training a convolutional network to pilot a simulated car on the GPU, than for impactful work at my actual job, which was a lot less challenging.
I guess hiring is just incredibly noisy, and I think companies could really get far hiring less than the best people, and just squeezing good quality work out of them (I believe Amazon is known for this).
Obviously OpenAI should not hire subpar people lol, they should keep doing what works for them, just grumbling loudly here.
2 replies →
It's both things. Offering more money increases the inbound applicants dramatically. You want this because top talent is disproportionately likely to have better options and may have a high floor. This does, however, create a non-trivial filtering problem, and one that is hard to scale because recruiters can't reliably differentiate real talent from good talkers. The key here is to make sure your best people are active engaged in recruiting and to minimize the type 1 and type 2 errors in hiring. This is hard to sustain at scale, and I think places like Google and Facebook have been losing the battle on this for many years—also due to engineering brand dilution because they don't actually have that much impactful work relative to the headcount—but OpenAI is small and focused enough that they ought to be able to manage pretty effectively.
> but in principle isn't offering a lot of money a good way to attract talent?
offering a lot of money attracts everyone including in-demand talent.
offering less turns away in-demand talent.
not everyone. There are motivations other than money, and if someone has a dislike of Sam Altman for whatever reason, they could be the best person in the world for a job and still not want to work there. If some has moral or ethical reasons against something, they won't be swayed by money.
You also have to give them something interesting to do but that isn't exactly a problem for recruiting to develop leading/bleeding edge tech