Behind OpenAI's plan to make A.I. flow like electricity
nytimes.com
nytimes.com
Or he wants his suppliers to be a regulated utility while he sells 100% margin products on top of it?
Or he wants all the greater fools now wherever they are so he can get out before the collapse?
It feels like this is a case of following the money to understand the real goals, since it's unclear to me that AGI is that goal.
Ironically, much of the observed behaviour is instrumentally convergent for most of the suggested ultimate goals.
Trying to make a safe and aligned AGI or a statutory government monopoly would both get about half the things we've seen.
The other half is stuff which is collectively mutually exclusive on all goals, but humans aren't perfect logical spheres in a vacuum, so it could still be basically any of them.
at this point, the lofty goals are a distraction from celebrating the utility of what we have, incremental upgrades, and reduced resource usage.
>Or he wants his suppliers to be a regulated utility while he sells 100% margin products on top of it?
Near term yep - soon as regulatory capture drops. They'll make out like bandits.
>Or he wants all the greater fools now wherever they are so he can get out before the collapse?
Then this, by the time competition catches up anyway because there are no real moats here besides regulation
>Wait, he wants his open product to be like a utility with low profits for maximized reach?
Then this is what remains, in a sea of other providers.
Financially? Pump and Dump, baby. Though I reckon the end result will still be intelligence flowing like water.
> As the availability of electricity became more widespread, people found better ways of using it.
This is an analogy, not a formula. And a great aspiration for a company that until recently was at the forefront of its industry.
Electricity didn't need the scale sama is asking for thriving as a tech (obviously eventually it did explode in use cases with enough scale, but that's not my point). Scientists used to make electromagnetism experiment demonstrations in Academies with equipment that was not that expensive nor inaccesible for the time. You can even produce enough electricity with a potato to light a bulb.
A better technology for this analogy would be promoting on-device/on-site open SLMs, but that's not what OAI promotes, is it?
> Scientists used to make electromagnetism experiment demonstrations in Academies with equipment that was not that expensive nor inaccesible for the time.
Treat it as a perdonal opinion but I really think the analogy resists no analysis after situations like that. If it is closed (and that's what OAI proposes, they aren't advocating for open models but the contrary), it will never be like electricity.
When all else fails, pitch your idea to the right officials in the USG and you'll make bank. And if you do it right, you won't even be doing anything at all. There are companies getting 8-10 figures a year to fail at upgrading systems from DOS to literally anything not DOS. It's a very low bar, but a profitable one if you can pull it off.
It also had hundreds of contractors and employs 200+ full time employees. It consumes $40M worth of electricity each year and uses 1.7 million gallons of water each day for cooling.
So it sure doesn't seem like it unless we want to get into a debate around how much perceived useful data and value the NSA gets out of spying on Americans and foreign countries.
I read some article about how the federal retirement system is about up to $500m over the last 20 years on a failed outsourcing the digitization of their paper system. They are apparently not much closer to having a digital version than they were when they started something like 3-4 contractors ago. Specifically the storage hub in West Virginia.
There has to be some sort of grift in that kinda waste. How do that many companies not move the needle and get to keep the money?
There has been much scrutiny over how much the ArriveCAN app cost to develop and who was subcontracted for its development. Contracts show that the federal government will spend close to $54 million with 23 separate subcontractors. A Parliamentary committee ordered federal departments to submit contracting documents related to the app but have been told that the names of subcontractors cannot be released citing issues of confidentiality. In October 2022, two developers at two separate IT companies took part in a hackathon where they both developed duplicates of the ArriveCAN app in under two days, for an estimated cost of $250,000.
How the f can this even be an excuse? How can the government be confidential from itself? If so why has the person that allowed confidentiality in these contracts not been removed. Rhetorical of course, "The un-accountability machine" grift in action.
No one in program offices (the offices responsible for these system developments) has the proper expertise to judge IT/software contracts or the IT/software portion of contracts. And, they won't listen to you because they've got a contractor or two sitting on their shoulder whispering into their ear things that are very wrong (this got a Colonel I worked with in trouble once) and leads to very biased decision making, away from reality and in favor of the grifters.
This is a reliable thing. There are some really good people in gov't, but they don't seem to make their way to the program offices. So without expertise and good judgement, you get these 8-10 figure (or maybe worse) boondoggles.
Would love to hear OpenAI's explanation behind this line of thinking.
I should have a machine that peels my potatoes. I should have a machine that cooks me my favorite meal at the push of a button. I should have a machine that I can feed fabric into and it will sew me a shirt. All of these things are doable, but they don't make economic sense or they need to handle too many unforeseen cases.
A machine that sews shirts doesn't make sense to have at home, but if you're selling millions of shirts then it does. Same for all the rest.
A machine that runs all the cables etc needs to be too complicated to handle all the eventualities that there isn't enough scale (yet) to justify its existence. Basically, we don't build enough data centers.
Would have to be energy. Data centres have light human footprints. And Altman wants to fabricate the chips in the Middle East, not America [1].
[1] https://www.bloomberg.com/news/articles/2024-04-10/openai-s-...
e: https://www.theverge.com/2024/9/20/24249770/microsoft-three-...
There's all manner of insane, ambitious things being worked on in the open right now, if you are able to muster the discernment necessary to ignore the natural, unavoidable pile of grift that follows any new technology or hype cycle.
And in your opinion, these are...?
With AI, I don't even know where to start. Have you used the internet recently? We have multimodal transformer models which can input and output images, video and text. We can create music and images using diffusion techniques. We have world-class text generation systems, which of course still need a lot of infrastructural support, but the viability is already being proven. We have cutting-edge translation tools which blow anything previous out the water. We have code generation and introspection tools, which work well enough to majorly augment my engineering workflow. We are developing ways to "speak" with your data, turning natural language into advance query, analysis and synthesis tasks. We have neural radiance fields. Lip syncing and voice translation tech to automate localization in powerful new ways.
And we have open models! Anyone can dive in, fine-tune a model and get viable results today. We can already start creating new pipelines today, so that we aren't wasting our time doing that once models get better at not hallucinating. The individuals and organizations who wait for this new technology to be perfectly reliable will get left in the dust by those who took the chance and did the pioneering work.
This is still barely scratching the surface. It seems like there's at least one ground-breaking paper a month, often more than that. It's certainly easier to bash these technologies and communities from afar, but that attitude will only hinder you in the future once the tech has caught up. What we have already created through experimentation just since 2015 is stuff that I grew up reading about in science fiction.
Could we hit an AI winter instead? Will decentralization ultimately be a dead end? Maybe. But it would be considerate to not bash those who are spending time and money to find out, and to not let grifters whom they cannot control co-opt the space and the narrative around it.
Beautifully put, thanks for writing my thoughts out.
Music generation is just mangled music from some popular band with custom lyrics, how can you think that you are creating something here?
This is good enough for 90% of image use cases, which is mildly relevant filler between and next to text.
If an artist enjoys using a tool to make something, and their patron enjoys the creative output, do you or I have any right to judge or admonish their process? Visionaries see what is possible, they don't spend their time critiquing things that aren't perfect and instead recognize and engage with novel, paradigm-shifting tools. I understand not wanting to interact with low-effort VC-driven bullshit, but there's a lot of amazing work being done in the open by scientists and enthusiasts.
The endgame, after these tools are developed and ironed out, will be a wealth of highly creative content like we've never experienced. We can shit on these tools or grab a shovel and try to improve them, which is what a huge community of people is doing right now despite the criticism.
It is my understanding that artists up in arms over AI are merely a noisy minority.
Artists who couldn't be chuffed or who like AI and perhaps even use it as a tool are not per capita likely to scream quite as loudly about their positions or try as hard to dominate the narrative.
Reading the headlines 200+ years ago it would have been easy to assume that all weavers hated Jacquard looms, as well. Especially in light of the fuming Luddites that had a tendency to break into shops and smash them up.
But at the end of the day, a person really has to pick a side: Is AI imagery worthless slop or is it a dangerous force that will replace human artists? I'd suggest that anyone legitimately unable to compete against worthless slop must be vastly overestimating the quality of their own work.
You say it's more efficient but I think if artists were fairly compensated for their work, or at the very least had a say in the process, it would be too expensive or not have good enough output to compete. On top of that, it uses A LOT more energy - which should be enough to exclude it as an option considering the current trajectory of climate change.
Aside from that we need to consider the more philosophical question of whether we want to make creative dream jobs even more impossible to find. Should we degrade the public space even more with an avalanche of "good enough" imagery simply to improve margins for the executive class? Computers have already made every level of developed economies significantly more productive and yet individuals are worse off financially than they were in the 90s. All this pipedream of replacing workers with AI will do is relegate them to manual labour or lock them into some barely-enough universal basic income scheme.
[1] https://www.dacs.org.uk/news-events/artificial-intelligence-...
Additionally, as an artist myself I don't really much care what "the majority" of artists I think, I didn't become an artist to follow trends and outsource my own critical thinking. As a software engineer, I have a level of understanding behind these models that far surpasses the average artist. As a technological visionary, I understand where the technology will lead, as well as its inevitability. The cat's out of the bag. So I don't very much value the average uninformed opinion.
> Aside from that we need to consider the more philosophical question of whether we want to make creative dream jobs even more impossible to find
Technological progress always leads to the destruction and creation of jobs. People similarly bashed the loom, the printing press, calculators, computers, the internet, cars, planes, you name it. There's no benefit in being a Luddite. Intelligent and aware engineers and artists are incorporating these tools into their workflows today, or at least staying informed, so that they find themselves still employed in whatever future is ahead of us.
This isn't ludditism. Ethical ML artistic tools do exist but none of these prompt-based generators fall into that category. None of the tools you mentioned are directly equivalent because they don't depend on a collection of existing work to produce derivatives.
All I will say about the climate is that emissions have to peak next year, and be halved by 2030 to meet the Paris Agreement goal. I have to assume from your casual dismissal that you're not aware of the consequences of missing it.
A) The courts are deciding this now, and B) we already live in a sick reality distorted by the DMCA and we need a major copyright reform.
> I have to assume from your casual dismissal that you're not aware of the consequences of missing it.
You managed to completely ignore what I said about diminishing energy requirements for compute. You're arguing disingenuously and now you're attempting to insult my intelligence instead of focusing on the argument.
It's really telling that this is the typical fashion in which these argument play out. I have no interest in continuing a conversation like this if you can't make good-faith assumptions in your argument as per the Hacker News guidelines.
The reason we love them is the same reason we love art… To create something out of nothing with the resources at hand.
Most artist love and hate, xor love to hate while simultaneously hating to love ai.
hysterical. exact same shit as always, folks
In many ways, the amount of money being put into cryptocurrency is antithetical to its development
But still, there is a major difference between AI and Bitcoin. With AI, the more power we put in the more we get out. That might be in the form of more accurate output, longer output, or larger context windows. With Bitcoin, more power going into the calculations just makes the calculations harder so the amount of useful work stays exactly the same. The calculations themselves have no use other than to waste time and power.
only after they are already built
My assumption is that the data centers would mostly be staffed by those who will have to manually audit data going in and out of the LLMs on a daily basis. There would also be a need to generate/curate the test data that the LLMs will have train on. There is potential for half a million jobs, but is that what you would want to have human effort invested in, is the real question.
There will therefore be a sudden supply of half a million such factories.
Think the Matrix, only the futuristic artificial intelligence isn't harvesting human brainpower/heat -- instead most people work at power plants.
Not working for you? Be the first to try the new elixir.
What happens if giving to everyone and their mothers ends up in low usage? Are they going to blame scale again and ask for a Dyson Sphere?
And it's kind of funny to see how well MS has played that particular instrument. A partnership with Sam got them 49% ownership and got rid of the doomer faction. The 75% profit share makes any future investors look long and hard at investing into what looks to be a capital-intensive low-revenue business right now.
Which makes it likely that OAI will run out of money, and oops, let's see if there aren't suddenly more than 49% ownership. And MS has a technology where they were long behind. (My bet is that the majority ownership will happen this funding round, because Sam is desperate)
The only way that doesn't happen is if Sam manages a massive investment round and finds a path to profit before the money's burned up. And the beauty of it is that the early 49% ownership means they only need to make a comparatively tiny investment this round to still be a majority owner in the company, with a 75% revenue share until their money is fully recouped. And the other investors get to bear the majority of the risk for the much larger funding round.
This are all utterly logical plays if you are willing to accept Sam is somebody who can make things move, loves gambling, and is a narcissist.
I really do think he's being played by a virtuoso, and it gives me great joy to watch that unfold.
Reminds me of a quote I saved from Kevin Kelly's book The Inevitable, which was published in 2016. I saved it because it sounded so absurd at the time. To be clear, it sort of still does sound absurd to me, who remain moderately skeptical about AI (though admittedly less so than in 2016, largely because of the following sentence). What's very serious is how much money and old-fashioned brainpower is going into making this future real. I don't know whether the metaphor was arrived at independently, derives from the same source, or is just an obvious way of thinking about the world when you've drunk the right flavor of Kool-Aid.
> Amid all this activity, of a picture of our AI future is coming into view, and it is not the HAL 9000—a discrete machine animated by a charismatic (yet potentially homicidal) humanlike consciousness—or a Singularitan rapture of superintelligence. The AI on the horizon looks more like Amazon Web Services—cheap, reliable, industrial-grade digital smartness running behind everything, and almost invisible except when it blinks off. This common utility will serve you as much IQ as you want but no more than you need. You’ll simply plug into the grid and get AI as if it was electricity. It will enliven inert objects, much as electricity did more than a century past. Three generations ago, many a tinkerer struck it rich by taking a tool and making an electric version. Take a manual pump; electrify it. Find a hand-wringer washer; electrify it. The entrepreneurs didn’t need to generate the electricity; they brought it from the grid and used it to automate the previously manual. Now everything that we formerly electrified we will cognify. There is almost nothing we can think of that cannot be made new, different, or more valuable by infusing it with some extra IQ. In fact, the business plans of the next 10,000 startups are easy to forecast: Take X and add AI. Find something that can be made better by adding online smartness to it.
Reality today seems to be anyone who's getting decent results out of ML is doing something slightly different according to the specifics of their domain. You can't take a chatbot LLM and expect it to predict the weather, architect a building, drive a car, and fold laundry.
https://news.ycombinator.com/item?id=38311509
Pure evil.
Whatever else he has done, he ranks super-super-low on the empathy scale...
https://www.lesswrong.com/posts/QDczBduZorG4dxZiW/sam-altman...
Search for "car payments". It's heartbreaking :'(
If I ran a company, I would not hire this man.
Shit
Edit: what is this lesswrong site about?
It's about philosophy. Or, at least that's the closest I can place it.
It has also been very interested on futurism, AI and the Singularity on the past.
This is very heartwarming.
Makes sense. I'd consider Ackman issuing comments on politics off topic for a hedge fund manager. But I suppose he's technically on topic with a narrow view.
> Altman however, dodges questions all the time, with vague corporate speak and generalities
He has a documented history of dishonesty, correct?
Yodkowsky shouldn’t be taken seriously. Especially after him saying that any person or country working on banned AI research should have their data centers destroyed via an airstrike.wonder if this guy considers stuxnet a wmd
If you have a massive warehouse to train an AI that, IDK folds proteins to develop a bioweapon or something, then you should not expect everyone else to sit on their hands while you plan to hold the entire world hostage.
Stuxnet was a program that targeted specifically SCADA controllers used in uranium enrichment facilities. It is not a doomsday device, it is a targeted counter-doomsday device. And while the USA has an imperfect track record with respect to "WMD"s/Proliferation, stuxnet is something that I approve of as an american.
That has been possible for a long time. The only thing needed was raw compute power to simulate it. Not sure how AI research suddenly makes itself worthy of a airstrike based on clearly unproven scenarios. What makes you think a rogue nation will sink ridiculous amounts of money building the power , cooling infrastructure, acquiring the necessary computing hardware and knowledge to build an hypothetical tool that will aid them in making a WMD - when they could just create tanks of sarin gas or anthrax for a fraction of the cost plus manpower that could be delivered to the target in multiple ways? A rogue nation having the capability to build AI datacenters, will also have the ability to do simple risk/benefit analysis.
I can’t think of many examples where he said something in an interview and it really challenged or surprised me. By contrast, any random hour with Lisa Su and you will come away quite impressed and easily understand why she is her company’s leader.
Plenty of smart money behind Altman, though, so maybe he shows up stronger behind the scenes.
[Edit: added "aloof" due to perceived aloofness.]
I don't think anyone doubts Sam's ability to charm people and tell a good story. The graver concern is "should we trust him"? My gut says NO.
“I LOVE THIS FOUNDER,” typed one partner.
“I am a 10 out of 10,” pinged another.
“YES!!!” exclaimed a third.
What Sequoia was reacting to was the scale of SBF’s vision....We were incredibly impressed, Bailhe says. “It was one of those your-hair-is-blown-back type of meetings.”
This is 'smart money' in reference to Sam Bankman Fried.
A common sentiment about SpaceX and Musk is that the employees also believe in the mission of making space accessible and going to Mars. Musk has been consistent on that being SpaceX's eventual goal since founding it, and he stuck through it even when SpaceX was one failure away from missing payroll. All of his early employees still publicly support this mission, even after retiring or moving on to found their own space companies.
Altman comes off as a "say anything to keep the money flowing" type guy even to his own employees. The fractured leadership all seems to suggest that he can't even manage to convince the people who work with him the most that he actually believes in his oft-repeated lofty goals regarding AGI.
I have a fond fantasy that those folks are laughing their asses off whenever we call the software stuff “high technology”.
Some people like to say that everything is harder in game development, but I'd say that it's even more so in semiconductor systems engineering. Tracing problems through every layer of the factory is like playing Elden Ring without a monitor. All you get is a guy in Korea who speaks very bad English to describe the scene for you. I can guarantee there is nowhere you will find more cursed problems. Imagine enduring a root cause analysis that winds up attributing a yield issue to fertilizer being spread on a field 20 miles away.
Downright harmful in some cases.
the "muh job" angle is well and good, but if this really convinces you, i got a bridge to sell. maybe sam thinks we all are into making podcasts now.
If AI fails to deliver its promises or fails to do so in the short timeframes implicitly proposed by the ecosystem, I expect dark times in terms of growth for the tech industry at least for a while.
Anecdote: As a on-off tech shopper dealing with some big providers from time to time, I can sniff the reek of desperation when trying to sell GenAI stuff. And that's weird.
I’m kind of curious where it all leads to. There’s a non zero chance it’s amazing given what is in my hands today is already amazing compared to what was in my hands 3 years ago.
I don’t think the choice is like AI or curing cancer. I think it’s more like Doge coin and meme stocks or whatever finance fad.
And profit!
What would be a concrete example of this? Has Salesforce revenue increased with their AI offerings? Has Microsofts?
* Capture the market (like Uber), by doing whatever it takes to be successful (e.g. offer your services at 1/3rd of what it costs) -- flush the equivalent of several countries' GDP down the toilet, if necessary.
* Wait for everyone to use only your service (because it's like water??)
* Raise rates (hopefully, everyone is "locked-in" at this point)
* Profit!
Also, it has almost no barriers to entry (AFAIK, closing a deal with NVidia is the largest one right now), so you can't do step #3.
Plus, we're talking about hyperscalers/big techs here, that would mean a commercial attrition war where OAI would be competing with actors such as Meta giving it for free from already existing platforms that have different main income sources (ig, fb, whatsapp). Good luck with that.
[1] Comments on Ask HN: What have you built with LLMs?
Writing of LLMs a being useless isn't wrong and unproductive, but assuming that they are universally applicable, or that they can function without supervision is also very wrong and potentially dangerous.