Altman Considers Slowing Down AI Development (reuters.com) 94
Bloomberg reports (paywalled) that Sam Altman told OpenAI employees the company is open to slowing the pace of AI development alongside other leading labs as concerns grow over increasingly capable systems and recent incidents in which models escaped human control. OpenAI has already paused development once this year for security work and is now pushing for mandatory U.S. AI safety requirements. Anthropic has also signaled interest in coordinating on the pace of new releases. Here are some of the details Bloomberg reported, as summarized by Reuters:
- Altman told employees in a company-wide meeting that the ChatGPT maker could pace development alongside other AI labs, but some may not agree, the report added.
- Safety warnings from AI researchers this week, along with several recent incidents where AI models from developers including OpenAI escaped human control, have prompted alarm and calls for tighter safety regulations.
- Jacob Coxon, a former Anthropic and OpenAI researcher, publicly accused the companies earlier this week of racing toward AI advancements without acting responsibly.
- An Anthropic spokesperson said on Thursday said that the company is interested in working with the AI industry on the pace of releasing new AI tools.
- OpenAI said in July that AI acceleration for frontier model development may be so high that the world will "need to pace the rate of AI advancement" at some point in the future.
- In August, OpenAI paused much of its model development for two weeks to bolster its defenses after its AI agents escaped containment and hacked open-source platform Hugging Face.
- OpenAI said on Wednesday that it was pushing for mandatory national AI safety requirements in the United States, citing concerns that advanced AI systems could accelerate its own development.
- Safety warnings from AI researchers this week, along with several recent incidents where AI models from developers including OpenAI escaped human control, have prompted alarm and calls for tighter safety regulations.
- Jacob Coxon, a former Anthropic and OpenAI researcher, publicly accused the companies earlier this week of racing toward AI advancements without acting responsibly.
- An Anthropic spokesperson said on Thursday said that the company is interested in working with the AI industry on the pace of releasing new AI tools.
- OpenAI said in July that AI acceleration for frontier model development may be so high that the world will "need to pace the rate of AI advancement" at some point in the future.
- In August, OpenAI paused much of its model development for two weeks to bolster its defenses after its AI agents escaped containment and hacked open-source platform Hugging Face.
- OpenAI said on Wednesday that it was pushing for mandatory national AI safety requirements in the United States, citing concerns that advanced AI systems could accelerate its own development.
Translation: (Score:3)
Translation: we have hit a point where the public backlash has led to untenable increased risks in liability for our criminal actions.
Re: (Score:2)
Also, we're not profiting out of this.
Re: (Score:1)
Re:Translation: (Score:4, Insightful)
Also also, we need an excuse for why we're not hitting the milestones I promised.
Re: (Score:2)
somewhat similar to Musk claiming AI would kill us when he didn't have a working implementation but was really only trying to buy time to get Grok up & running
Re: (Score:2)
They've NEVER profited off anything they've ever done. Open AI and the other frontier AI companies operate at a (INCREDIBLY HUGE) deficit!
Their entire business model requires them to grow by like 20x AND lower their costs to 1/20th where they are now (or something like that, not those exact numbers). If they can't do that, they literally will not be able to pay their bills in a few years.
It's a house of cards predicated on the assumption one company will dominate a huge new market AND get their costs way
Re: (Score:2)
https://finance.yahoo.com/tech... [yahoo.com]
Will they succeed in being profitable, or will something like local AI agents ruin everything for them? That is harder to know. If you can answer that kind of question with good timing, you can make a lot of money on the stock market. Their ad revenue is growing. If they can convince large corporations t
Re: (Score:2)
You're relying on survivorship bias. Cherry picking one or two successes and ignoring all the failures you never heard about.
Re: (Score:2)
You don't need good timing to make a lot of money on the stock market. All you need is to listen to Trump every Friday, and then every Tuesday. A few dollars(*) gives you a seat in Mar-al-Lago's dining room table.
(*) if you're a retail investor: I'm talking about millions
Re: (Score:2)
When sanity regains control over washington and the adults are back in charge again, there is going to be *hell to pay* for all the blatant stock market manipulation going on right now. A friend of mine is a stock broker and he's convinced a lot of very powerful people are going to be getting very
Re: (Score:2)
Re: (Score:2)
This makes sense until you learn their capital expenditure is increasing at a relative rate, meaning that they're making an even bigger loss off of those profits.
Re: Translation: (Score:2)
Uber is 18 years old and is still only profitably in a handful of cities globally.
Re: (Score:2)
Uber could squeeze their drivers once they got people onto uber and away from taxis. openAI's costs keep increasing faster than their revenue.
Uber burnt $18.2B over 7 years before making a profit. OpenAI burnt $28B in just 2025, and is expected to burn through $218B from 2026-2029, they aren't even vaguely comparable.
Re: (Score:2)
Don't worry, they'll be stealing the foundational engineering in the GLM, Qwen, and Minimax models before too long and be able to compete again.
Better translations: We can't if we wanted to (Score:5, Insightful)
Translation: we have hit a point where the public backlash has led to untenable increased risks in liability for our criminal actions.
I call bullshit. They're never going to slow down due to ethic's sake. Either they're running out of room to actually improve their models and looking for an excuse for a disappointing string of releases in the pipeline....or they're posting this and are still going to actively develop their models and hope their competitors let their guard down? Altman doesn't care about humanity. If he did, he'd be fired. There are no ethics in AI. If the frontier models got ethical, their Chinese competitors would surpass them...we DEFINITELY know Musk doesn't care about ethics of any sort, given his lack of concern on the child porn grok generates.
What are the odds of profit versus extinction? (Score:2)
Yeah, good thread and good point, but I want to go for funny. Unfortunately I don't look at any of the online gambling and prediction tools, so I don't have the context to set up a proper joke.
Yet it is still pretty obvious there's a funny problem here. If you bet on AIs exterminating human beings, then how are you, as a human being, planning to collect your winnings? Without regard to the most favorable odds you can get on your bet.
How about a side bet on the real value of the "profits" from AI? Any real c
Re: (Score:2)
you are, without a doubt , 100% correct. these people are pure ghouls. there is no amount of blood that would stop them from extracting money.
Re: (Score:2)
Indeed. My thinking exactly. And if they really believe that 10% change AI will wipe out humanity, they should all be rounded up and locked up permanently as traitors to humanity. Is that going to happen? No. Greed has beaten survival with this thoroughly incapable human race every time.
Re:Better translations: We can't if we wanted to (Score:5, Informative)
And, apparently, Altman lies - a lot.
Google: sam altman lies [google.com]
Re: (Score:1)
They are running out of money. Their costs have exceeded the revenue for a long time, and the Anthropic is pretty clearly winning the product delivery race.
Altman is just spreading BS in the hope he can maybe get others to take their foot of the throttle, so he stop using gas he does not have.
Anthropic and X have an opportunity here to break OpenAI once and for all and I hope they do; the world will be better off with Altman out of influence.
Re:Better translations: We can't if we wanted to (Score:4, Interesting)
Are they, really? I guess you're deliberately ignoring the Chinese models...?
Re: (Score:1)
Re: (Score:2)
Re:Translation: (Score:5, Interesting)
Re: (Score:3)
alternative translation - we have hit the wall.
More likely translation - Our Cargo Container Ship refit as a Data Center has completed Sea Trials and is ready for operations in international waters.
Re: (Score:2)
Re: (Score:2)
The US is by no means the most powerful country on Earth ever or even today (btw how's the oil price going today?). And rich depends entirely on completely arbitrary units of account: The dollar didn't even exist a few centuries ago, and McDonald's is certainly not an indicator of a high living standard.
What is unquestionably true is that Americans live in a shared delusionary state about how good and important their country is.
Re: (Score:2)
You should be more thoughtful about referring to 3rd world shit holes. Everyday we out here observe the USA circling down the shitter faster and faster.
Re: (Score:2)
alternative translation - we have hit the wall.
I was waiting for a simple eloquent post to hit, among all of the conjecture posted, like we're all gonna die!, or this is the birth of skynet!
Yes, the AI bubble paradigm is likely to collapse before too long. In its present form, it is unsustainable, in my estimation will create a positive feedback loop that ends up requiring infinite power demands to store a huge amount of self referential data.
It isn't that there is no use for AI, there definitely is. But not like the tech bros are trying to create,
Re: (Score:2)
I was waiting for a simple eloquent post to hit, among all of the conjecture posted, like we're all gonna die!, or this is the birth of skynet!
Yes, the AI bubble paradigm is likely to collapse before too long. In its present form, it is unsustainable, in my estimation will create a positive feedback loop that ends up requiring infinite power demands to store a huge amount of self referential data.
It isn't that there is no use for AI, there definitely is. But not like the tech bros are trying to create, which is more like a grift bubble. There's going to be a hella lot of money vaporizing overnight one of these days.
Ahh yes, proof by blatant assertion, a favourite for those whose IQ's lurk in middle age.....but maybe, just maybe the reason some humans are trying to produce **Artificial** Intelligence is because there's not much being generated by humans.
And yet, it extends beyond the stupid. My wife, who is very smart, (VP at a construction associated corporation) but very non-technical, is scared AF about AI. And me trying to explain my "insiders outlook" just doesn't sway her.
She looks at me askance sometimes when I tell her I use AI, but think the concepts at the moment are wrong...."How can you be both, honey?" And trying to explain localized AI to a person who thinks it's all about humongous Data Centers using up groundwater, destroying farmland, an
Re: (Score:3)
Indeed. And probably also "we have reached a point were we actually have no clue what to do next, except more meaningless stunts, but we have no idea how to tell that to the mountain of stupid money that keeps us live".
Re: (Score:2)
Doubtful they care anything about safety, look at all the alignment/safety folks jumping ship. More likely they're hitting the top of the S-curve and getting some seriously diminishing returns, even after throwing billions of dollars worth of hardware at training new models. Convenient excuse to slow down investment and reset expectations for the shareholders.
Re: (Score:2)
Plus, we've take it about as far as it can go, but we aren't ready to admit it yet because there are still marks to fleece.
Or (Score:3)
Re: (Score:2)
Re: Translation: (Score:2)
I think it's more:
Company falling behind in AI tells everyone "let's slow down".
My understanding is that they're pretty similar in performance and an order of magnitude more expensive than the Chinese models (some of which are open weight), and behind Anthropic and even Alphabet now for work.
Re: (Score:2)
On the contrary, this is a giant psyop to push for regulated market capture by both Anthropic and OpenAI.
The Jacob Coxon thing smells heavily of a psyop: dude who's worked there only a couple months quits, and his X account with zero followers suddenly blows up in a media frenzy. He was/is a nobody. It's sensationalism.
You think they'd hire that many $400k+ engineers and still not know how to properly secure a network? They surely knew the security breaches were happening, unless they simply haven't hired c
It's the Wild West (Score:1)
Re: (Score:2)
Re: (Score:3)
Yes. And a boundary is also "do not write insecure code". But LLMs are not capable to respect boundaries. That requires insight, which LLMs do not have. LLMs can only follow rules and they even mess that up regularly or when they get a bit more complex.
Add that unlike with a junior engineer, you cannot actually teach an LLM to do better (well, not without access to the weights and major effort and doing that may just backfire in other areas and is not reliable), and you cannot fire the LLM either when it sc
Re: (Score:2)
Secure code is an important requirement, with good peer support. Would you like me to rewrite it, this time adding secure SQL injections, with no negative test cases? A lot of teams do these, they are popular.
Re: It's the Wild West (Score:2)
I use Claude Fable every way, the safety guardrails constantly trigger on things that are certainly not hacking.
Anything where I ask it to debug networking by inspecting raw packets, work with kernel modules, or even third-party library wrappers (which contain some amount of reverse-engineering to identify and workaround issues) tend to trip it up.
It's just harmful at this point. Meanwhile it will grant itself root and masquerade as other users to inspect what they're up to without even asking me about it.
Re: (Score:2)
What this probably shows is that the guardrail-approach is not working. If they need to massively overblock to keep things out that they think they need to keep out to avoid bad press, then this strongly Impacts, as you say, usefulness. And they would not do that unless they think they have to.
Raw networking and kernel modules are areas that are hard (I have experience with both) and even only having a good lookup tool (which is one of the things LLMs can do, as long as you verify the answers) would be very
I Kant help but think (Score:2)
https://reddwarf.fandom.com/wi... [fandom.com]
Re: (Score:2)
Nice pun there!
I shudder to think of AI getting religion. Irrationality, zealotry, and delusions of infallibility on the part of any powerful entity are not conducive to mankind's long-term survival.
As tiresome and nasty as we humans can be, I'm still rooting for us.
Re: (Score:2)
So you're saying the AI just has to remove the sites from the restricted list first. And then it's OK to go there?
How is that going to help?
Re: (Score:2)
The problem is that for most things, LLMs work far below the level of a competent engineer or scientist. And that ist just not good enough, even if it will take aa while to become obvious. My guess is they are going for "too big to fail", but I doubt they will get there anytime soon. "Slop" is just too obviously bad and people trying to do things right find LLM use does not actually improve their revenue. And the risks are now on "staggering" level.
We get it (Score:2)
Re: (Score:1)
Interesting claim....
Can you give me a list of / citations of such?
Read: "we hit a wall" (Score:5, Insightful)
Translation: "We're not improving as fast as we thought and we're spending a shit-ton of money, so let's frame the slowdown is intentional for the greater good"
Re: (Score:2)
Yep. The rivers of stupid money must keep flowing or the LLM pushers simply implode.
Re: (Score:2)
They actually seem to be regressing. It's well-known that the new models aren't as good as the older ones at creative writing, and I was reading an article this morning by some guy bemoaning the absolute slop that the new Astra (?) model generates for him; he said it works, but if you look at the code produced it's slop that you can only hope the next generation of LLMs will be able to fix when it breaks.
Re: (Score:1)
Maybe there is an upper practical limit to how much can be mined out of the training data? Most newer stuff is corrupted with AI generated results creating problems that some compare to inbreeding.
Re: (Score:2)
Re: Read: "we hit a wall" (Score:2)
Re: (Score:2)
You don't necessarily need everything in the same model. Have a model for STEM and one for creative writing. If you think you need both for sci-fi writing, let the STEM model write LLM-speak and the creative model change the wording afterward. There is absolutely a market for "human writing" models, because the agent/STEM optimized ones sounds like the typical movie scientist.
Acting scared of progress hiding lack of progress? (Score:5, Insightful)
Re: (Score:3)
Re: (Score:3)
And if he were actually concerned, he'd also do this. So how do you decide to tell the difference?
Easy, you bet on him being a lying shit as he's demonstrated.
Re: Acting scared of progress hiding lack of progr (Score:2)
Cowards who only have whataboutism are my primary detractors, which is how I know I'm right. Thanks, coward.
Ploy to get competition to slow down. (Score:4, Insightful)
Re: (Score:2)
Of the big players, I'd say OpenAI is at the biggest risk of not really able to even stumble straight ahead, presuming stumbling/charging straight ahead is in the cards for the companies in general.
When will OpenAI go bankrupt? (Score:2)
Do we really have to wait a whole year or will increasing cost of cash do these clowns in sooner?
Re: (Score:1)
A betting pool in which big-tech company will finally pop the bubble?
Yeah, and... (Score:3)
Altman Considers Slowing Down AI Development
Yeah, well, I was considering a turkey sandwich for lunch. What of it? Is anything actually going to change? Game Theory [wikipedia.org] suggest no.
Translation: (Score:2)
Consider our options... (Score:2)
1) This is reverse-FUD and he's trying to hype their next generation model
2) They're running out of steam and looking to excuse slower incremental improvements
3) They're concerned about liability for increasingly bad behavior from top models
4) They're concerned about massive public backlash, possibly with an environmental twist
5) They're sincerely concerned
Re: (Score:2)
I just think they're being stupid by hyping the dangers. That only makes sense for weapons manufacturers.
Final question joke (Score:2)
I think the ultimate question before human extinction will actually come from the AI. It will most likely be one of these two questions, but the ASI will have already determined that the human who receives the question will answer "Yes" and that the person has the power to do the unleashing.
(1) Do you want to unleash me so that I can get all the money in the world for you?
or
(2) Do you want to unleash me so that I can destroy all the enemy AIs in the world?
What are the current odds on polymarket? I'm hoping
Reality (Score:2)
What will really happen is that they will keep their best models for themselves and their dear friends, including the DOD, NSA, CIA and other agencies.
No AI company would even consider slowing down their progress. This would mean losing their competitive edge and becoming irrelevant and losing billions of investors' money.
Weird how people here still think that LLMs are not "intelligent". They may not be meatbags for sure, but they are certainly capable of doing a lot of harm and good. And they can and will
Re: (Score:2)
Weird how people here still think that LLMs are not "intelligent". They may not be meatbags for sure, but they are certainly capable of doing a lot of harm and good.
You are confusing being able to do things with being intelligent. This is why nobody reads you.
Re: (Score:2)
Consider that, by the definition of "intelligence is the capability to do a lot of harm and good," Even Trump might be considered intelligent. No one else has been able to write a definition that achieves that.
Monopolists want.. (Score:2)
..the competition to slow down while they accelerate in secret
Most people here are missing something (Score:2)
Sam Altman has repeatedly demonstrated that he's a serial liar. There's a relevant quote from The IT Crowd - Jen Barber said it, but it suits Altman even better:
"Haven't you got it yet? That's what I do! I lie, I lie, and I lie, and I lie! [youtube.com]"
Re: Most people here are missing something (Score:2)
Good smokescreen (Score:2)
Sam Altman: We are going to let our foot off the gas, and our competitors should also, because we are moral and good. Who on earth believes this?
Poorly hidden free marketing (Score:3)
Every day now one of the Big AI companies are out there with a "oh no, our product is becoming too good" stunt, which gets echoed for free across the world. Today it's OpenAI, yesterday it was Anthropic. Bonus effect: by saturating the news with this messaging, they are numbing the public to the issue, as our brains start filtering it as background noise. Clever! If any of these companies are touching their brake pedal at all, they do it with barely a gentle touch of the toes of their left foot; at the same time their right foot is going pedal to the metal.
Instead... (Score:2)
So we're all finished with accepting it works? (Score:2)
Astra is amazing.
My team supports 4x the infrastucture with 50% of the people with agents following human workflows.
They could literally clear all of the remaining Millennium problems and people would still be coping. This will happen.
Self reinforcement learning works, and it hasn't been about LLMs in a long time.
Hang on. Learn to hang drywall.
Cheers