sarker
competency crisis actor
Suddenly I cannot remember the color of your eyes
Or the things we said as we stood together for the last time
User ID: 636
Ok, so can you please make your point plainly?
My point is that the talent in this field is basically exclusively clustered in the Bay Area and China, as evidenced by the inability of anybody else to produce any model worth a damn. It really doesn't matter that top Israelis join the military - their top guys aren't very good, or they don't have enough of them.
The question I asked is how many people are earning a million dollars salary in these AI research positions. Are you saying it's "thousands"?
Yes! 4300 people are going to make millions of dollars on the Anthropic IPO alone. This doesn't count any cash income and it's one lab.
Why? Isn't there lots and lots of government land out West which are already used for secret military projects? Is there a deed for Area 51?
Sure, and since data centers are visible from space we can figure out how much capacity is going up on government land from open source Intel. It's not a lot compared to what the hyperscalers are doing (150 GW of planned capacity).
Again, can you please just make your point plainly?
The point is that you don't have any argument except "they could be doing this, despite the fact that we see none of the evidence we'd expect to see if they were doing it".
Can we agree that they have a lot of skilled people and a lot of money?
I'm deeply uncertain about how many skilled AI researchers are employed by the government.
Maybe not the best, but they are hiring a lot of excellent people. For example a starting Assistant US Attorney might make $100,000 a year A starting US Foreign Service officer makes significantly less than an Assistant US Attorney. An analyst at the federal reserve bank might make $80k a year. There are lots of very smart capable people working in all of these jobs. Agreed?
The median lawyer only makes about $160k. $100k is really not a bad starting salary. If you're an AI researcher at the NSA, your top salary potential is half or less the starting comp in the private sector. It's just not a comparable situation.
And of course in Israel, it really is the best people given that there is mandatory military service and their computer unit is highly prestigious. Agreed?
I'm agnostic about whether the best Israelis join the government, but my point is that the best Israelis just aren't very good, or there aren't enough good ones to matter, as evidenced by the paucity of successful Israeli tech companies.
And would you also agree that it's possible to staff an AI Manhattan Project with excellent people, even if they are not the best?
Certainly. With sufficient investment and aggressive recruiting such a program might be able to reach parity with last year's Chinese models.
No idea. I don't even know what a "frontier model" is though.
The answer is "zero".
How many of these jobs are there?
There's probably thousands of L6 or higher engineers across the frontier labs. More employees at lower levels still earning a lot. Literally nobody is earning just $200k, which is the upper limit of an NSA salary.
Does it include land and land improvement costs?
It does. The NSA would also have to purchase land to build data centers though.
Maybe the hypothetical AI Manhattan Project is funded through the US Military budget. Can we agree that this is a good deal more than Google's "capex"?
Sure, and maybe there's an invisible dragon in my garage.
The reason I am skeptical is pretty straightforward: Defense planners around the world surely realize that these AI systems have immense military implications.
They might, but that doesn't mean that they can do anything about it.
Also, do you agree with me that there are a lot of smart and talented and capable people working for the government who could be making a lot more in the private sector?
There's some for sure, but in fields where you can actually make a decent living in the private sector the government is not hiring the best.
There’s a pattern of indie musicians coming from elite families.
Why not:
-
Kids of rich families don't need to hold down a job
-
Making music is somewhat more respectable than being a layabout
-
"Indie rock" is close to the lowest common musical denominator for a particular age democratic
So if you're a mostly untalented failson scion it's basically a great option.
In Israel, it's pretty well known that most of the very best young people join Unit 8200 each year.
Cool. How many frontier models have they produced? You're right, that's not fair, they could be classified.
How many frontier models has the Israeli private sector produced?
How many of these millions of dollars a year jobs are there anyway?
Even a non-top researcher, even a software engineer, say an L6 or so, can earn a million dollars a year at a frontier lab.
If you say so -- what I heard, I heard from what I consider to be a reliable source. Perhaps "capex" includes more than just computer equipment?
It's more than computer equipment in the sense that a DC chiller isn't technically computer equipment, but it's equipment you need to run the computers and it's a lot more specific to "computer equipment" than the entire budget for the NIP which is dwarfed by a single hyperscaler's capex.
I'm guessing you were told by a government employee who doesn't realize that it's not 1975 anymore - I think that really shows you the level of situational awareness in our government today.
For years, it's been well known that the biggest employer of mathematicians and the biggest purchaser of computer equipment in the US is the NSA.
Mathematicians can basically either work for the NSA or become an adjunct professor or high school teacher. Top AI researchers can either get millions of dollars a year from a frontier lab or $200K from the NSA. What do you think is the more likely choice?
As far as computer equipment purchases go, this is just false. Alphabet (alone!) is spending $200B on capex (alone!) this year while the entire National Intelligence Program budget is $90B. This claim was probably true during the cold war though.
How many captains of industry did our ancestors prosecute for adultery?
He's already married.
Mochas are my go to, but I found a coffee place near me that has a cubano which is also pretty good.
You can just get a prescription through e.g Amazon Pharmacy.
It seems like it's only a problem if you aren't a person of wealth. I have good credit, but unless I'm planning on buying a house I can't see it mattering in the near future.
I believe it does affect your credit once it goes to collections.
If you can afford to pay you will pay.
Is that really true? My impression is more like "the bill will eventually be sold to collectors for pennies on the dollar and then you can bargain with them to pay a few more pennies than they paid and it all goes away." Why is that wrong?
Jesus may not match your blood type.
The peak popularity is only somewhat greater than "gisele pelicot" though. I think there's a pretty simple explanation of the remaining gap.
Pelicot was front page on e.g. The Guardian, FT, etc. where the Cornell 7 were not. She also got a NYT magazine cover story.
I think part of the reason the Pelicot gang-rape case didn't get as much attention as it perhaps should have
Huh? How much more attention could something have gotten?
A model can trivially be trained on every math publication published in the last hundred years, while a human can see only the tiniest fraction of them.
The models already are trained on them, and probably have been since 2022.
All of these discussions cause people to reflect on the process of doing math. They’re like, “Oh, this thing can do this impressive stuff. What does that mean for us?” One of the people in that video highlights this quote: “good mathematicians prove theorems, great mathematicians come up with conjectures, and the greatest mathematicians come up with definitions.” That’s more or less exactly your framing here. We need the conjecture generator and then the definition generator. That’s the premium-tier mathematician.
What is it that causes current models to be great theorem provers and poor conjecture and definition makers? It's unclear, and perhaps this comment will look foolish in six months, but it's not as simple as "train on all the math papers bro".
It's cool but this is a flying car in the same way chickens can fly (please don't "well ackshuakly" me on this point). The "you could have cured cancer" crowd is also dumb. It's a parlor trick, though I don't begrudge Elon his parlor tricks. What would I actually have found impressive in a car demo? Probably not something easily demoed. How about a Bluetooth connection that has three nines of reliability?
If you decide to undereat (without chemical assistance*), your body will reduce your metabolic rate and make you hungrier
Do you believe that this process continues without bound? The more you cut, the more your metabolic rate drops in lock step?
Maneuvering is the leading cause of accidents.
Man emerged from a process of continual speciation, not from a sudden creation. At which point did a mere animal become ensouled? Who was the animal mother that bore a human child?
It's better not to think about this question, it might lead you to accord moral weight to animals, which would be inconvenient.
You can use a free model. Or pay a few dollars (is that really all your money?) and use a better model.
I don't understand your perspective. Either LLMs are useless, in which case, what are we even talking about in this thread, or they aren't, in which case putting your head in the sand is not an effective long term strategy. "LLMs are eating my field and they're a racket" does not seem like a coherent position.
seems extremely-similar to the proof case of why the halting problem's not always solvable
And yet it's trivially true that there are programs that can be proven to halt and entire programming languages with which you can write code that's proven to halt. Solving the halting problem is simply a question of choosing your tools; why not alignment?
- Prev
- Next

Sure. Just how you wouldn't be reading about the invisible dragon in my garage, if it did exist.
I don't know what you expect from me - I've provided yearly comp numbers showing that staff positions at Anthropic pay over a million, and there's a lot of people at staff level or close to it across the labs. The fact that a lot of people profiting aren't top researchers is the point, you don't even need to be a top researcher to make this kind of money and top researchers get a lot more. Or is the difficulty here in "comp" vs "salary"?
You are welcome to continue ignoring the comp disparity between the private and public sector. You are in good company!
More options
Context Copy link