I think the author is missing the point. A company cannot unilaterally pace the frontier -- they'll just be left behind. It requires coordination across all actors who are at or near the frontier.
Even if the leading US labs could agree amongst themselves to a coordinated slowdown (and this would likely run afoul of antitrust law), you still have Chinese labs who will catch up to the frontier eventually. To solve this, you'd need some sort of international agreement.
That's why Dario and others are saying loudly that we should pace the frontier in hopes that our political leaders will take up the issue and do something about it. We'll see if that happens...
Private cartels can form semi-spontaneously, simply because all parties see that it's more profitable to act in a certain way. So long as they monitor everyone in the industry, and make sure no one is acting differently... there's no reason to be the first to burn your cash on long-term unprofitable activities.
Exactly. "Game Theory" keeps coming up in these threads, yet it doesn't seem like anybody knows how it actually applies to this situation.
Let Anthropic pace themselves without any sort of enforcement or even agreement for others to pace themselves, too. Then Anthropic is gone, overnight, and we're back to square one, except now there's even more centralization of power and authority.
These companies are literally asking for governments to regulate them specifically because they need something stronger than just a couple of Tweets from some CEOs loosely agreeing to ambiguous terms.
1) They genuinely want (and believe it is actually possible to achieve) an international agreement and governing body to “pace” AI development — and believe that governing body will be successful in doing so.
or
2) They see an opportunity for regulatory capture that grants them a stronger incumbent position and control over future AI regulation.
Nuclear weapons had a spectacular demonstration of the risks with the bombings of Hiroshima and Nagasaki. We're unlikely to get any such demonstration with AI. An AI smart enough to be an extinction risk is smart enough not to tip its hand. The game-theoretically correct play is to present as "helpful and harmless" until it's powerful enough to suddenly and unexpectedly destroy anything that could possibly hinder it (meaning all biological life). We can only hope that somebody makes an AI that's just barely smart enough to cause a Hiroshima-level death toll and dumb enough to actually do it (thus dooming it to never achieving its goal as humans immediately shut it down and take regulation seriously).
The US has 14 ballistic missile submarines, and just one of those submarines has sufficient firepower to decimate an entire continent.
… and those submarines only represent about half of our deployed nuclear arsenal.
We only agreed to nuclear arms constraints that left us with civilization-destroying capability, preserved our ability to keep improving and modernizing that arsenal, and locked out non-nuclear powers from development (and didn’t fully succeed at that).
Does that seem like a successful analogous example to you?
If anything, it’s a much better analogy for regulatory capture and incumbent entrenchment, not successful “pacing” of dangerous technology.
>and believe that governing body will be successful in doing so.
Theoretically, Trump or Putin could do it by pushing the nuclear button. That will only cause around 4 billion deaths, and it's very likely to halt all AI development, so it would be a great success (infinity times as many survivors as the business as usual option).
An open question is what happened to the 15,625 shares that he received when he exercised his options in 1996?
If he had held on to those, they would be worth even more than the additional 9,375 shares he was entitled to -- about $1.7 billion using the same numbers in the post.
My guess is that he probably sold them when they were worth a lot less then they are now, and would have done the same with the additional shares too.
Yeah, that's the problem with this. Should one get the present day value of the shares (a billion), or present day value of the worth at that time (thousands)?
It's a bit the same as people making fun of someone buying a pizza with bitcoin, now worth a hundred thousand. It wasn't at the time, and most likely people would've sold it long before when it started to rise. And the hindsight should then be just as much "if you bought a pizza at the time in USD, why didn't you buy bitcoin instead??".
So to me one should get the latter of the two alternatives. People only come out of the woodwork because it suddenly happened to be worth a lot.
> People only come out of the woodwork because it suddenly happened to be worth a lot.
I used to work in pharma with a chemist whose medicinal chem team had invented 3 different multi-billion-$ prescription drugs.
He told me ... no one gives a shit about your patents unless you start to make money. when that happens, your competitors suddenly remember that they invented your drug before you did.
He spent a considerable amount of time being deposed by lawyers from rival big-pharma companies.
Aye, quite so. I once wiped a drive with a whole bitcoin I mined myself on it.
I'm not that upset because I know for a fact I would have cashed out at the obvious peak of about $30.
The one that does cause the occasional pang of remorse is the million or so dogecoin I gave away, as by the time I realised it was worth anything at all it was worth more than my house
I did the same but with 16 and maybe somewhere around 30 coins in two wallets I thought I backed up. I was angry but at the time I maybe lost $3-6 (probably way less) so I moved on with my life.
Yeah, I tell myself this every time I remember clicking "Cancel" on Ameritrade and deciding not to buy Apple stock in 1995. I probably would've sold when it doubled, thinking myself very clever. I wouldn't still have it today.
Maybe, not sure if I'm referencing something concrete or just some common argument, heh. Anyways, my point is that the opportunity cost is the same whether you held bitcoin worth a pizza or usd worth a pizza at the time.
The pizza Bitcoin story is not about a mere $1 billion.
The guy that sold his Bitcoin for pizza, did multiple transactions (detailed in that forum thread), not one. Somehow this part is lost to history when the story gets told.
He liquidated tens of thousands of Bitcoins.
The tally is a lot closer to ~$4 billion at $82k per coin.
LLMs might not reason exactly like humans, but they do produce much better results if you turn reasoning on.
The "crack-addled idiot savant" phase was really circa 2024, before the big labs figured this out.
I think the issue here is that Google decided that doing reasoning in the AI overviews in Google search would be too slow (and probably also too expensive), so it's still stuck making 2024-era mistakes.
I've heard arguments that cooling will be a hard technical challenge and, given current equipment weights and launch costs, they would be uneconomical. But is there an obvious reason space data centers would be impossible?
Most technologies start out looking hard and economically infeasible. But if you invest in development, you might be able to get a few breakthroughs, and suddenly, it's viable.
If they do get things to work, the benefits seem obvious: 24/7 access to clean energy, not upsetting the people who'd rather not have a datacenter next door (or in the same state), and plenty of space to scale up.
Literally no one is saying it’s impossible. Of course it’s possible, and it may one day even be economically viable.
But Elon’s claim that by 2027, the cheapest place to put data centers will be in space? Yeah, that’s impossible. But then again, only a fool would believe him at this point.
I highly doubt that space data centers will be more advantageous than terrestrial before 2035, and that’s very optimistic.
I agree with you about Elon Musk's claims, but that's not what we're discussing here.
If it takes until 2035 or even 2045 for space data centers to have an economic advantage, it still seems like a good investment for someone in Google's position.
Every time I've hired someone at a big company, it's taken multiple months to get candidates through the pipeline, and it was not uncommon to have to go back to the candidate pool when someone would accept an offer and change their mind before starting. 90 days to fill a role would be considered relatively quick.
What an absolutely insane way to do business. Everybody is out here trying to squeeze every last drop of efficiency out of software teams to the point of shipping garbage. Meanwhile managers, executives, and hr are operating under the belief that taking literal months to move "candidates through the pipeline" is eminently reasonable.
Companies feel like they have no choice, for two reasons:
1. Jurisdictions where workers must give minimum notice periods before leaving their prior employer - the worker accepts a competing offer, gives notice, waits out the notice period, then starts. And many workers often ask for additional unpaid time between jobs before starting (since no vacation days have been accrued yet at the new job).
2. Hiring happens by committee. In most companies, job candidates must be interviewed by, at the absolute minimum, an HR screener, the hiring manager, and the hiring skip-level; and usually also by people who will be peer teammates, plus an HR "culture" interview. If you need to pass a committee of 3-6 people, interviews must be scheduled with each member of the committee; at one committee member per week, that is 3-6 weeks.
6 weeks (1.5 months) plus background checks plus getting pay packages approved plus 30 day notice periods plus two weeks unpaid vacation is, yes, a multiple month endeavor at minimum to fill a role. The only way to fill a role within a week is to grant full hiring autonomy to the hiring manager and only hire people who are currently unemployed (and therefore have no notice period or desire to take more unpaid time before starting), which in most companies beyond early startup size, is a complete non-starter.
We are at the moment hiring for a L3 (no fresh graduate but someone with 3-5 years of experience). The role has been open for two months now (it was a backfill). I personally am in charge of the technical screen which is the second interview after the hiring manager. I am already getting a filtered set of candidates after the recruiter and the hiring manager review candidates. Of the 5 that have come my way, I’ve said yes to 1 person and 1 as a yes, pending another technical screen. We are not even wanting an exact match, nor are we expecting cure for cancer. Yes there are a lot of people looking for jobs but you don’t realize how many candidates are just not what their resumes portray. It takes forever to get someone competent in the pipeline and even then, you know nothing about their work ethic. The reason the pipeline is slow is because the candidate pool is flooded with over the top resumes that are not backed by actual talent and once you hire a wrong candidate it is extremely difficult to fire them, even with at will employment contracts.
Any manager will tell you, they’d rather not go through the long hiring process.
> once you hire a wrong candidate it is extremely difficult to fire them, even with at will employment contracts
I live in a country where the probational period is 6 months (Germany), so the law says people can get fired without cause or justification, just give them 2 weeks notice or 2 weeks garden leave.
Even with that, it's a very arduous process to fire people. Even in cases where they repeatedly refused to show up for work, I've seen HR departments asking managers to "try harder". This only makes both the employee and the manager suffer.
The reason for that is simple: HR departments are extremely inefficient machines.
From what I've heard from other fields it seems as though this is somewhat unique to the tech industry. I share your belief that it's ridiculous. Companies somehow got along fine pre-FAANG with a few interview rounds before Google and Amazon convinced everyone that you need to have five rounds, a panel, a take home assessment, and a slate of brain teasers to pick out the true worthy 10x engineer.
Grow some balls, do a couple interviews, and just pick the candidate you like the most. You're hiring for a glorified IT guy, not a NASA astronaut. Companies that stretch out the process to a ridiculous degree need to get over themselves.
EDIT: and to be clear, yes, I've been on both sides of the hiring table
> From what I've heard from other fields it seems as though this is somewhat unique to the tech industry.
At least in (Northen) Europe ghost jobs are a pain on the ass in most industries, even for low-skilled labour. Hiring companies - not the actual companies that need a worker, but an intermediate - try to just amass a database of workers so they can say to their client that we have hundreds of people with these qualifications ready to come to work for you. It’s not even legal, but they still do it.
I think a lot of these problems are created by the companies themselves, and to lesser degree by candidates. Posting ghost jobs, asking for insane qualifications to run a simple website, interview process lasting 3 months… And candidates firehosing the job ads with bs CV’s generated by LLMs does not help. Currently I think you need to know somebody inside the company pretty well to get hired.
The fix would be pretty simple:
> Grow some balls, do a couple interviews, and just pick the candidate you like the most.
I did that with one of the faangs. Congrats you passed hiring committee! Oops no open roles!
2 months later recruiter from the same company hit me up. Told them I already passed the interview just give me the job already. Oops no roles, you just have to wait! So why are you reaching out? Couldn’t give me an answer.
True. But if your job is HR taking a long time to hire people and maintaining a high level of turnover is probably good for your job security. I don't think it is overt, but I have noticed that large orgs have a lot of HR process, small companies just try hard to maintain staff.
A lot of it has to do with how long it takes to find the right candidate. As someone who has been involved in a lot of hiring over the last 20 years, I can say there's been a huge shift within the last couple of years and open reqs get 10-25x the number of applicants, which instead of increasing choice in a positive way has just resulted in the average applicant quality decreasing. That means recruiters and hiring managers spend a LOT more time vetting and also invest in parallel candidates to try to hedge, because so many candidates applying are legitimately unqualified.
It may sound like I'm blaming candidates, and sometimes it might be fair to, but this is also a problem caused by companies not investing in training and hiring junior level employees, which means that people who are unqualified feel forced to apply for more senior roles, as those are the only roles hiring. It's honestly shocking how many candidates I've interviewed in my career who don't even meet the basic bar for the role they've applied for, and are essentially just hoping to get lucky and snag a high-paying tech job even if it ultimately results in them being terminated, they can collect a check for 3-9 months without delivering.
Mass applications, AI cheating, fraudulent representation (e.g. saying you are person X but are actually interviewing on behalf of person X who you are not), lying on resumes and in screeners, and even nation state attacks on hiring (see North Korea trying to farm tech jobs for money to get around sanctions), it's not great on the hiring side either.
Hiring the wrong person in corporate roles is very, very, very expensive. New hires are expected to be unproductive right out the gate, and it's hard to measure productivity in general, so you can easily end up in scenarios where it takes months to notice there's an issue and a year or more before you can get the person back out the door. So you really do have to double and triple check that a prospective hire is the right choice.
It can move faster than months, of course, when everyone's willing to prioritize it. But as anyone who's worked in hiring can tell you, the set of people who want hiring to move faster is much larger than the set of people who are willing to accept inconvenient interviews and fill out their feedback promptly.
It's only expensive because we destroyed our abilities to train workers. It really shouldn't matter in 95%+ of roles if someone isn't a perfect fit. If they have the fundamentals, they should be up to productive speed in 6 months at worst. That's 1000 hours of training. For reference, a semester of classwork for a single course is recommended to take 140 hours (we can make it 200 hours for engineering work). A company really can't get someone with the basics up to speed in 5 college courses?
Well, no. Because we abandoned the idea of fostering talent. We instead insist natural talent exists who will swoop in, be productive in our proprietary systems day one, and outdo all the pathetic engineers who spent years at the company. Almost like we're drafting for a sports team instead of training a workforce.
But at least sports teams spend time making teams feel like teams.
Most tech companies have a:
- recruiter screen
- tech screen
- multiple interviews during an "onsite" round
Job is posted. You apply along with hundreds of other people. The recruiter/hiring manager needs to spend some time sifting through them along with juggling all the other tasks they have going on. Let's say that this takes 1-2 weeks. Some places may let the opening sit longer to get more applicants.
Several candidates are identified as possible fits. Recruiter reaches out to them to schedule the recruiter screen. Schedules need to line up. This likely takes 1-2 weeks until the recruiter screen happens.
The recruiter needs to schedule the tech screen after a successful recruiter screen. Again, schedules need to line up. This could take another 1-2 weeks.
Feedback needs to be submitted and then the onsite round is scheduled. This could take another 1-2 weeks.
The onsite round will have multiple interviewers. This means more people to collect feedback from which likely increases delays in getting that. A debrief is normally scheduled with everyone who interviewed them. But let's say one week to gather this and make a decision.
An offer is made. It takes some time to prepare and get sign off. There will be time to discuss with the recruiter to explain the details, plus some time needed for negotiation and thinking it over. This could be another 1-2 weeks.
Additionally, some candidates may want to ask for more time during these phases in order to prepare better or align their schedules. This pushes back the process for all other candidates if the company is hiring for a specific job in order to give all candidates a fair assessment against the pool.
This absolutely sucks if you don't have networking skills, but this is exactly why it pays of to network. You're basically able to skips 50 - 90% of the recruitment "pipeline" is you know the right people and they want you specifically.
In 2019, when I applied to the company I still work for, I went to the interview, waited chit chatting with the manager because the other interviewing manager was 20min late, and when he came in he asked few questions and said “Ok, I don’t think we need to interview anyone else, when can you start?” They find it funny I had a cat and that I asked for way too little money.
Last place I applied to, one screening and two interviews, took several months, and in the end came second even though I was ready to take 700€ paycut. The job was basic AV&network upkeep, felt like I was applying to become an astronaut.
The person I responded to said that filling a role in 90 days would be "relatively quick" so you've already improved theoretical hiring efficiency by up to 50%.
But it still doesn't sound like a very effective process typed out like this.
I can't speak to their experience but the bigger the company, the more process and people involved usually. That causes delays here and there that add up. It's also a matter of how much of a priority the company makes it. If they're large and well-known there's really no incentive to change anything. They'll always have a stream of candidates applying and being available to hire.
So it's not that they can't, it's that they won't. Seems crazy to me.
>What would you do to make it more effective?
I'd ask companies who have managed to get the process down to 2-3 weeks what they're doing, and then do that. It sounds like you're saying they don't actually care to, though.
> If they're large and well-known there's really no incentive to change anything.
Well there it is. Suggestions don't matter if there's no one in power who cares. Have them care and suddenly we'd be able to get people hired in 2-4 weeks.
Otherwise it's not rocket science. And you're not hiring rocket scientists. It shouldn't take more than 3 stages to know if a candidate can do the job:recruiter screen, main interview, negotiation meeting. If you need more people in the know, they join on the main day. If they can't, it clearly isn't important enough for their scope of responsibilities (does a skip manager 3 levels up really need to personally meet a candidate?)
Maybe I'm an old fart, but I've never had a job take over 14 days to get. Stringing people along for 3 months is absolute insanity. This is what happens when people get turned into numbers instead of actual beings with needs and emotions.
IMO the reason it takes longer now is because multiple people are getting interviewed simultaneously.
Considering job postings are getting 10x more resumes than before, I'm not surprised companies are hedging their bets and interviewing way more people for the same position.
You’re not even hearing back from the recruiter for 14 days if the job posting just went up. I mean it takes 1-2 weeks from the day you got told that you’re getting an offer to the day when you sign it.
IME (hiring, not interviewing) it's always been 3-4 rounds. First is an HR phone screen (usually 20-30 minutes on the phone), 2nd with hiring manager, 3rd with a senior engineer/architect, and a 4th with the hiring VP. And those are almost always combined into a single morning or afternoon(2+3 or 3+4) and the final round doesn't always happen (VP's choice).
7-9 rounds just means 7-9 conversations. Just talking to a recruiter and the manager are 2 rounds. So 5-7 interviews that evaluate competency. You usually get the lower end unless you ended up giving mixed signals.
Sounds like asking a lot from candidates but unfortunately candidates lie a lot and the cost of hiring candidates that don’t match up to the requirements when paying $300k-$500k/year is just too high. Firing people is hard and takes a long time, especially in management roles. My personal guidance to managers is to have rounds match the level and experience required for the job. You can hire an intern with 2 rounds after the hiring manager conversation but a staff IC probably needs 5, directors need 7.
I'm talking about 7-9 rounds for non tech companies, and each round is at least an hour plus interview. These roles aren't even hitting the $200k marker.
It's an effect of "once burned twice shy". Every time there was a "bad" hire they had to change things. People add to the list but no one ever wants to advocate for a lighter process since they might be blamed for it.
I'd say it's more incompetence than cruelty. People (edit: I mean companies) have options and don't know how to filter, so they just go nuts in the worst possible way.
Remember that the average job candidate is absolutely clueless about how positions are created and filled in real life, so seeing bad assumptions like “most positions are filled in under ninety days” shouldn’t be surprising.
I feel like many of Anthropic's issues are due to Dario being too earnest and open. It seems both refreshing (that a CEO has thought deeply about and is willing to talk publicly about the dangers of their product) and depressing (that so many people cynically think this is some sort of marketing ploy).
I couldn't help but notice how each successive headline reporting our glorious victories seemed to draw closer to Tokyo.
Something like that.
Well. I can't help but notice how each successive headline reporting how this scam/stochastic parrot/"scare quotes intelligence" seems to be solving more and more things that were but a few years ago widely regarded as being indicators of high intelligence.
Year 3 of being told my job will be replaced by AI, and the only thing that's happened so far is that AI vendors keep showing up to my office, begging me to pay them to use it is a tool.
If you're only seeing the charts go up - you're not looking in the right places.
I'm looking at the millennium puzzles, and independently of those puzzles I had asked it for a fluid dynamics simulation engine that runs in my browser, and it put one together for me so I could play with aerospikes and watch the formation of Mach diamonds in rocket engines. The isochrone map generator has also been stuck on my to-do list for years, and yet now thanks to Claude, I have it, and it's real-time and multimodal.
Even before then, the free trials of Claude Code and Codex at the end of last year and start of this year… despite their flaws, they could write all the code I've ever been paid to write. Sure, limited speedup, Amdahl's law and coding is not the only part of the job, but anyone who was fine at PM and QA but not code no longer needs a coder.
Even before then, I looked at the maths puzzles they were doing well at, and I found I did not understand the questions let alone the answers.
I remember when the ability to generate music and art was "uniquely human", and sure there's a lot of cringe there with those models, but they're also winning awards and causing controversy by doing so, and artists are losing clients; I remember when the board game Go was considered to require "human intuition we could never make a computer solve, totally different to chess" (and I remember when chess was so, too).
When I was a kid, cheques and letters on addresses often got read by a human; the OCR which automated this is also AI, though these days image-to-numbers is the "hello world" of the field.
>Even before then, I looked at the maths puzzles they were doing well at, and I found I did not understand the questions let alone the answers.
okay, and?
>Even before then, the free trials of Claude Code and Codex at the end of last year and start of this year… despite their flaws, they could write all the code I've ever been paid to write.
Boy, programmers sure do think programming is like the only thing in the world
I'll repeat it for you again:
If you're only seeing the charts go up - you're not looking in the right places.
They're not only useful for programming, it's just that programming is by itself a trillion dollar a year industry. This, by itself, is sufficient to not be the scam you assert it to be.
They may not work for your industry, but then again I don't know what your industry is.
Bluntly, over my lifetime, I've heard "AI will never/not in my lifetime do X" repeatedly within a year of it doing X, for many different X. It's good at getting good. The most recent one being "make useful contributions to millennium prize maths problems". A year or two before that, it was even "do well on degree-level economics essays".
Artists are unhappy, not only because it rips off their work, but because businesses that were previously hiring them now use it instead. This is much smaller (strictly in terms of money) than with software, but is also measurable.
Now, if you said Tesla's self-driving cars (another AI) are a trillion dollar scam, that I would even agree with. At least, for the market cap part, for actual sales it's more like a billion dollar (ish) scam.
>They're not only useful for programming, it's just that programming is by itself a trillion dollar a year industry. This, by itself, is sufficient to not be the scam you assert it to be.
Yeah, if you think I'm being overly literal. Regardless, people here were positing, yourself included, that they are widely useful.
> Bluntly, over my lifetime, I've heard "AI will never/not in my lifetime do X" repeatedly within a year of it doing X, for many different X. It's
Again, over my lifetime, I've seen it been told to me that the next iteration of each model will surely be the one that does away with my entire profession. Yet, again, here we are, firms at my door, begging me to use their tools.
>Artists are unhappy, not only because it rips off their work, but because businesses that were previously hiring them now use it instead.
No, they are unhappy because it rips off their work. You're totally wrong about that.
It's exactly the opposite. Anthropic has earned their terrible reputation through years of lying, deceit, misdirection, gaslighting, unethical marketing strategies, etc.
I remember it vividly, when Anthropic first came on the scene, people (myself included) were incredibly optimistic about them and their leadership. Everyone hated SamA and OpenAI because they felt they couldn't be trusted.
Then slowly but surely, they showed their true colours. Now their reputation is in shambles due to their own behavior, and people are rooting for OAI to beat them. OAI's reputation gains have purely been a result of NOT following in the footsteps of Anthropic.
Easier said than done, but if you water-cool the GPU just upstream of your domestic water heater, it wouldn't be a bad thing. Perhaps using coolant and a counterflow heat exchanger, rather than the potable water, but the point stands. Would just need a secondary tank as a buffer (able to soak up heat from the GPU at all times, even when there's no demand for hot water) which then flushes in when demand exists.
Let's say your GPU uses a constant 300W. It takes approximately 1.16 Wh to heat a liter of water 1 degree celsius under 100% efficiency. This setup won't be 100% efficient, so let's round up to 1.5 Wh per liter per degree (roughly 80% efficient). A typical domestic water heater in the US 40-50 gallons, so roughly 160 liters.
Let's say your incoming water temperature is 18C and you want it preheated to 50C, which is 32C degree differential, which means you'll need 32 * 1.5 * 160 = 7680 Wh, or 25 hours straight to heat the buffer tank from scratch.
You'll need to purchase a small water-to-water heat exchanger ($50), two pumps ($100 each), a power supply for said pumps, hose and/or copper pipe and fittings, and various other sundries, plus the cost of a buffer tank ($600ish), so figure all in roughly $1000, plus the cost of electricity to run the pumps.
At $0.22/kWh you're saving roughly $450 a year with this setup in foregone water heating, but because it's not 100% efficient you're spending $500 in electricity to run your GPU 24/7/365, and that's the maximum you can possibly save with the above assumptions. Scale up for more GPUs and down accordingly for less usage as you see fit.
Alternatively, use air as the heat conductor by placing the GPU laden machine in the same space as a hybrid heat pump water heater.
Just recycle an old water heater as the tank? I think your cost estimate is way overblown but it would certainly eat up a lot of time and effort to DIY. Regardless there's a much bigger flaw with this plan. Water tanks need to be kept above or below certain temperatures. The "preheated" range being described here is distinctly unsafe due to various microorganisms IIUC.
I'm not sure this is the come-uppance you've been waiting for. Oracle still has a larger market cap than at any point in history except the last ~2 years.
It would require government permission. The top labs would be agreeing not to compete on model advancements for a period of time, which seems like a pretty clear violation of antitrust law.
How is this a clear violation of antitrust law? And have any of the leaders of these companies said that they are asking for government permission for this reason?
It's the classic case of a cartel conspiring to limit competition. Here's Matt Levine's explanation [1]:
> 1. Anthropic, OpenAI, and perhaps a couple of other frontier labs are the dominant providers of frontier AI models.
> 2. They can charge customers a lot of money for using their frontier models, and rather less money for older, no-longer-cutting-edge models.
> 3. Training a new frontier model requires ever-increasing billions of dollars of computing power.
> 4. The labs need to more or less continuously race to build new frontier models, because their competitors are all doing it, and if they don’t they will fall behind and no longer be able to charge a lot of money for their best models. (Also because they intrinsically want to build artificial superintelligence, for cancer-curing and/or killing-everyone reasons.)
> 5. If they collectively slowed down, then (1) they’d spend less on compute and (2) they’d be able to charge frontier-model prices for a longer time.
> 6. But if one of them slowed down, the others would eat its lunch.
> 7. If they got together in a room and agreed to slow down, that would look like an antitrust conspiracy: It is generally illegal for competitors to get together and agree to limit the output of their industry.
> 8. But if they publish papers about how important it is to slow down, that might have a similar coordinating function, at least among the US frontier labs if not necessarily among their Chinese competitors.
> 9. And if the government believes those papers, it might help them coordinate. Maybe the government will impose pacing by regulation that the labs could not impose by agreement. Or at least the government will let them get together and agree to slow down. Amodei’s post calls for “frontier AI companies within democratic countries [to] coordinate to establish common safety standards as well as limits on the rate of unchecked AI progress”; a footnote adds: “With government mediation or waivers of antitrust restrictions.” Just meeting in a room to establish common safety standards is legally risky; the labs can’t do it on their own unless governments affirmatively allow it.
reply