I work at OpenAI. Software developers are not obsoleted by Codex or Claude Code, nor will they be soon.
For our teams, Codex is a massive productivity booster that actually increases the value of each dev. If you check our hiring page, you’ll see we are still hiring aggressively. Our ambitions are bigger than our current workforce, and we continue to pay top dollar for talented devs who want to join us in transforming how silicon chips provide value to humans.
Akin to how compilers reduced the demand for assembly but increased the demand for software engineering, I see Codex reducing the demand for hand-typed code but increasing the demand for software engineering. Codex can read and write code faster than you or me, but it still lacks a lot of intelligence and wisdom and context to do whole jobs autonomously.
This seems like a reasonable take. Maybe you could inform your CEO, the media and influencer sycophants, the tech companies that are laying off tens of thousands of developers while mandating the use of your company's tool, and everyone else responsible for us being inundated with outlandish claims that software engineering is dead on a literally daily basis. Hey, while I'm asking for wishes that won't be granted, maybe get people in your company to stop thinking they're so important that it's okay to buy 40% of the world's RAM supply with borrowed money, making it cost 4.5x as much for the rest of us?
In the text, we did share one hallucination benchmark: Claim-level errors fell by 33% and responses with an error fell by 18%, on a set of error-prone ChatGPT prompts we collected (though of course the rate will vary a lot across different types of prompts).
Hallucinations are the #1 problem with language models and we are working hard to keep bringing the rate down.
Yeah, long context vs compaction is always an interesting tradeoff. More information isn't always better for LLMs, as each token adds distraction, cost, and latency. There's no single optimum for all use cases.
For Codex, we're making 1M context experimentally available, but we're not making it the default experience for everyone, as from our testing we think that shorter context plus compaction works best for most people. If anyone here wants to try out 1M, you can do so by overriding `model_context_window` and `model_auto_compact_token_limit`.
Curious to hear if people have use cases where they find 1M works much better!
> Curious to hear if people have use cases where they find 1M works much better!
Reverse engineering [1]. When decompiling a bunch of code and tracing functionality, it's really easy to fill up the context window with irrelevant noise and compaction generally causes it to lose the plot entirely and have to start almost from scratch.
(Side note, are there any OpenAI programs to get free tokens/Max to test this kind of stuff?)
Do you maybe want to give us users some hints on what to compact and throw away? In codex CLI maybe you can create a visual tool that I can see and quickly check mark things I want to discard.
Sometimes I’m exploring some topic and that exploration is not useful but only the summary.
Also, you could use the best guess and cli could tell me that this is what it wants to compact and I can tweak its suggestion in natural language.
Context is going to be super important because it is the primary constraint. It would be nice to have serious granular support.
I too tried Codex and found it similarly hard to control over long contexts. It ended up coding an app that spit out millions of tiny files which were technically smaller than the original files it was supposed to optimize, except due to there being millions of them, actual hard drive usage was 18x larger. It seemed to work well until a certain point, and I suspect that point was context window overflow / compaction. Happy to provide you with the full session if it helps.
I’ll give Codex another shot with 1M. It just seemed like cperciva’s case and my own might be similar in that once the context window overflows (or refuses to fill) Codex seems to lose something essential, whereas Claude keeps it. What that thing is, I have no idea, but I’m hoping longer context will preserve it.
Yeah, I would definitely characterize it as an instruction following problem. After a few more round trips I got it to admit that "my earlier passes leaned heavily on build/tests + targeted reads, which can miss many “deep” bugs that only show up under specific conditions or with careful semantic review" and then asking it to "Please do a careful semantic review of files, one by one." started it on actually reviewing code.
Mind you, the bugs it reported were mostly bogus. But at least I was eventually able to convince it to try.
It occurred to me that searching 196 .c files was a context window issue, but maybe there’s something else going on. Either way, Codex could behave better.
Haha. This was the second time in like a year that I’ve posted a Twitter link, and the second time someone complained. Okay, I’ll try to remove those before posting, and I’ll edit this one out.
Feels like a losing battle, but hey, the audience is usually right.
I'm sorry, but it's my pet peeve. If you're on iOS/macOS I built a 100% free and privacy-friendly app to get rid of tracking parameters from hundreds of different websites, not just X/Twitter.
This is great! I have been meaning to implement this sort of thing in my existing Shortcuts flow but I see you already support it in Shortcuts! Thank you for this!
So what is your motivation for doing this, incidentally? Can you be explicit about it? I am genuinely curious.
Especially when it’s to the point of, you know, nagging/policing people to do it the way you’d prefer, when you could just redirect your router requests from x.com to xcancel.com
It's not particularly about x.com, hundreds of site like x, youtube, facebook, linkedin, tiktok etc surreptitious add tracking parameters to their links. The iOS Messages app even hides these tracking parameters. I don't like being surreptitiously tracked online and judging by the success of my free app, there are millions of people like me.
so, since these companies have to comply with removing PII, is the worst thing that could happen to me, that I get ads that are more likely to be interesting to me?
i’m not being facetious, honest question, especially considering ads are the only thing paying these people these days
The worst thing that could happen is that you get caught in some government dragnet based on your historical viewing data and get disappeared because (as is the nature of dragnet searches) no matter how innocent you are you still look guilty.
Who has to comply with removing PII? Your profile, yours, mapped to a special snowflake ID, is packaged and sold across a network of 2500 - 4000 buyers, including in particular those that clean, tie (a surprisingly small footprint turns into its own "natural primary key"), qualify, and sell on to agencies. No step in this is illegal.
my first and last name is already a "natural primary key" (every single google result of Peter Marreck is me), so I've already had to give that up a long time ago. So nothing new is lost I guess?
The more data they have on you, the more valuable that data is to a third party. So they sell your data to someone else, who then phones you based on your known deep interest in <whatever it was that tracked you>. Or spams you. Or messages you. Or whatever method they think will most get your attention.
If you don't give them that information, they can't sell it, and the buyers won't annoy you.
It's not that the ads you get are more interesting, it's that you get more ads because they think they know more about you.
Helpful type of nagging for me. Most here would agree they are not a positive aspect of the modern digital experience, calling it out gently without hostility is not bad. It might not be quite self policing but some of that with good reason is not bad for healthy communities IMO.
It's funny that the context window size is such a thing still. Like the whole LLM 'thing' is compression. Why can't we figure out some equally brilliant way of handling context besides just storing text somewhere and feeding it to the llm? RAG is the best attempt so far. We need something like a dynamic in flight llm/data structure being generated from the context that the agent can query as it goes.
My favorite solution is a lower parameter 5 layer model trained on the data that acts as a local compression and response, a neurocortext layer wrapped around any large persistent data you have to interact with and ...... maybe also a specialist tool that spins up which is built with that data in mind but is deterministic in it's approach- sort of a just-in-time index or adaptive indexing
That’s actually a pretty cool idea. When I think about my internal mental model of a codebase I’m working on it’s definitely a compacted lossy thing that evolves as I learn more.
Personally what I am more interested about is effective context window. I find that when using codex 5.2 high, I preferred to start compaction at around 50% of the context window because I noticed degradation at around that point. Though as of a bout a month ago that point is now below that which is great. Anyways, I feel that I will not be using that 1 million context at all in 5.4 but if the effective window is something like 400k context, that by itself is already a huge win. That means longer sessions before compaction and the agent can keep working on complex stuff for longer. But then there is the issue of intelligence of 5.4. If its as good as 5.2 high I am a happy camper, I found 5.3 anything... lacking personally.
Not sure how accurate this is, but found contextarena benchmarks today when I had the same question.
It appears only gemini has actual context == effective context from these. Although, I wasn't able to test this neither in gemini cli, nor antigravity with my pro subscription because, well, it appears nobody actually uses these tools at Google.
That's an interesting point regarding context Vs. compaction. If that's viewed as the best strategy, I'd hope we would see more tools around compaction than just "I'll compact what I want, brace yourselves" without warning.
Like, I'd love an optional pre-compaction step, "I need to compact, here is a high level list of my context + size, what should I junk?" Or similar.
This is exactly how it should work. I imagine it as a tree view showing both full and summarized token counts at each level, so you can immediately see what’s taking up space and what you’d gain by compacting it.
The agent could pre-select what it thinks is worth keeping, but you’d still have full control to override it. Each chunk could have three states: drop it, keep a summarized version, or keep the full history.
That way you stay in control of both the context budget and the level of detail the agent operates with.
I do find it really interesting that more coding agents don't have this as an toggleable feature, sometimes you really need this level of control to get useful capability
Yep; I've actually had entire jobs essentially fail due to a bad compaction. It lost key context, and it completely altered the trajectory.
I'm now more careful, using tracking files to try to keep it aligned, but more control over compaction regardless would be highly welcomed. You don't ALWAYS need that level of control, but when you do, you do.
Have you tried writing that as a skill? Compaction is just a prompt with a convenient UI to keep you in the same tab. There's no reason you can't ask the model to do that yourself and start a new conversation. You can look up Claude's /compact definition, for reference.
However, in some harnesses the model is given access to the old chat log/"memories", so you'd need a way to provide that. You could compromise by running /compact and pasting the output from your own summarizer (that you ran first, obviously).
Frontend work with large component libraries. When I'm refactoring shared design system components, things like a token system that touches 80+ files, compaction tends to lose the thread on which downstream components have already been updated vs which still need changes. It ends up re-doing work or missing things silently.
The model holds "what has been updated" well at the start of a session. After compaction, it reconstructs from summaries, and that reconstruction is lossy exactly where precision matters most: tracking partially-complete cross-file operations.
1M context isn't about reading more, it's about not forgetting what you already did halfway through.
What needs to be an option is to allow complete and then compact and if needed go into the 1m version. That way you can get the most out of the shorter window but in the case where it just couldn't finish and compact in time it will (at cost) go over. I wonder how many tokens are actually left at the end of compaction on average. I know there have been many times where I likely needed just another 10-20k and a better stopping point would have been there.
This observation makes sense, because all models currently probably use some kind of a sparse attention architecture.
So the closer the two related pieces of information are to each other in the input context, the larger the chance their relationship will be preserved.
I really don't have any numbers to back this up. But it feels like the sweet spot is around ~500k context size. Anything larger then that, you usually have scoping issues, trying to do too much at the same time, or having having issues with the quality of what's in the context at all.
For me, I would say speed (not just time to first token, but a complete generation) is more important then going for a larger context size.
I have found a bigger context window qute useful when trying to make sense of larger codebases. Generating documentation on how different components interact is better than nothing, especially if the code has poor test coverage.
I've also had it succeed in attempts to identify some non-trivial bugs that spanned multiple modules.
context distillation mostly. Agents tend to report success too early if they find something close to what they need for the task. If you are able to shove it in a 1M context, it's impossible for them to give up looking, it's in the context. But for actual implementation, it's not useful at all. They get derailed with too long of a context.
Yeah, for a while ChatGPT Plus has been powered by two series of models under the hood.
One series is the Instant series, which is faster and more tuned to ChatGPT, but less accurate.
The second series is the Thinking series, which is more accurate and more tuned to professional knowledge work, but slower (because it uses more reasoning tokens).
We'd also prefer to have simple experience with just one option, but picking just one would pull back the pareto frontier for some group of people/preferences. So for now we continue to serve two models, with manual control for people who want to choose and an imperfect auto switcher for people who don't want to be bothered. Could change down the road - we'll see.
By the way, I imagine you know this, but the product split is not obvious, even to my 20-something kids that are Plus subscribers - I saw one of them chatting with the instant model recently and I was like "No!! Never do that!!" and they did not understand they were getting the (I'm sorry to say) much less capable model.
I think it's confusing enough it's a brand harm. I offer no solutions, unfortunately. I guess you could do a little posthoc analysis for plus subscribers on up and determine if they'd benefit from default Thinking mode; that could be done relatively cheaply at low utilization times. But maybe you need this to keep utilization where it's at -- either way, I think it ends up meaning my kids prefer Claude. Which is fine; they wouldn't prefer Haiku if it was the default, but they don't get Haiku, they get Sonnet or Opus.
I agree -- we're on the ChatGPT Enterprise plan at work and every time someone complains about it screwing up a task it turns out they were using the instant model. There needs to be a way to disable it at the bare minimum.
You could perhaps show the "instant" reply right away and provide a button labeled "Think longer and give me a better answer" that starts the thinking model and eventually replaces the answer.
For this to work well, the instant reply must be truly instant and the button must always be visible and at the same position in the screen (i.e. either at the top or bottom, of the answer, scrolling such that it is also at the top or bottom of the screen), and once the thinking answer is displayed, there should be a small icon button to show the previous instant answer.
That's assuming that the instant answer is even directionally correct. A misleading instant answer could pollute the context and lead the thinking model astray.
Can the context of the pre-revision, Instant response be simply be discarded -- or forked or branched or [insert appropriate nomenclature here] -- instead of being included as potential poison?
(It seems absurd that to consider that there may be no undo button that the machine can push.)
For those who are unaware, this is exactly what Grok does. The default is an auto mode, then when you ask a question it starts researching (which is visible to the user) and if it's using the expert mode but you don't really need all that jazz, it has a "Quick Answer" button right above the prom entry field, and if it's using a "Quick Answer" mode then it has "Expert" button and the same place, and you are able to toggle between them mid answer and it will adjust the model (or model parameters, I'm not sure how it works under the hood).
It's pretty good with the auto chooser, but I appreciate the manual choice available so in-your-face and especially not having it restart the query completely but rather convert the output to either Quick or Expert.
This is on the Web UI, can't speak for other harnesses. I do find that it's quite good with the citations and has a fairly generous free tier, even on Expert mode. (As for who sits at the top, I am indeed put off by Musk's clear interference in several cases involving Grok, nor do my personal values align with the majority of his, but today's Grok is definitely less MechaHitler and more reliable than it was before.)
Thanks for clarifying! I guess the default for most users is going to be to use the router / auto switcher which is fine since most people won't change the default.
Just noting that I'm not against differentiation in products, but it gets very confusing for users when there's too many options (in the case of the consumer ChatGPT at least this is still more limited than in pre-GPT 5 days). The issue is that there's differentiation at what I pay monthly (free vs plus vs pro) and also at the model layer - which essentially becomes this matrix of different options / limits per model (and we're not even getting into capabilities).
For someone who uses codex as well, there are 5 models there when I use /model (on Plus plan, spark is only available for Pro plan users), limits also tied to my same consumer ChatGPT plan.
I imagine the model differentiation is only going to get worse as well since with more fine tuned use cases, there will be many different models (ie health care answers, etc.) - is it really on the user to figure out what to use? The only saving grace is that it's not as bad as Intel or AMD cpu naming schemes / cloud provider instance naming, but that's a very low bar.
Auto will never work, because for the exact same prompt sometimes you want a quick answer because it's not something very important to you, and sometimes you want the answer to be as accurate as possible, even if you have to wait 10 minutes.
In my case it would be more useful to have a slider of how much I'm willing to wait. For example instant, or think up to 1 minute, or think up to 15 minutes.
That's pretty close to what they have. They just named them Instant, Thinking (Standard), and Thinking (Extended), and they're discrete presets instead of a slider.
Yeah I use that, but it's not really a solution that allows to only have auto. It doesn't help when it chooses Instant instead of Thinking, and it's also much slower than using Instant outright because the Skip button doesn't immediately show, and it's generally slow to restart.
I've long suspected as much, but I always found the API model name <-> ChatGPT UI selector <-> actual model used correspondence very confusing, and whether I was actually switching models or just some parameters of the harness/model invocation.
> One series is the Instant series, which is faster and more tuned to ChatGPT, but less accurate.
That's putting it mildly. In my experience, the "instant/chat" model is absolute slop tier, while the "thinking" one is genuinely useful and also has a much more palatable tone (even for things not really requiring a lot of thought).
Fortunately, the latter clearly identifies itself with an absurd amout of emoji reminiscent of other early chatbots that shall not be named, so I know how to detect and avoid it.
Before GPT-5 was launched, and after sama had said they would unify the ordinary and reasoning models, I think we all expected more than an (auto-)switcher, we expected some small innovation (smaller than the ordinary-to-reasoning one, but still a significant one) that would make both kinds of replies be in a way generated by a single model (don't know exactly how, I expected OpenAI to surprise us with something that would feel obvious in retrospect).
The model doesn't even need to be exposed in the UI. Let the user specify "use model foobar-4" or "use a coding model" or "use a middle-tier attorney model".
VIM does this well: no UI, magic incantations to use features.
Forgiveness but while you're here can you look into why the Notion connector in chat doesn't have the capability to write pages but the MCP (which I use via Codex) can? it looks like it's entirely possible, just mostly a missing action in the connector.
I'm an OpenAI employee and I'll go out on a limb with a public comment. I agree AI shouldn't be used for mass surveillance or autonomous weapons. I also think Anthropic has been treated terribly and has acted admirably. My understanding is that the OpenAI deal disallows domestic mass surveillance and autonomous weapons, and that OpenAI is asking for the same terms for other AI companies (so that we can continue competing on the basis of differing services and not differing scruples). Given this understanding, I don't see why I should quit. If it turns out that the deal is being misdescribed or that it won't be enforced, I can see why I should quit, but so far I haven't seen any evidence that's the case.
Respectfully, it's very hard to see how anyone could look at what just happened and come to the conclusion that one company ends up classed a "supply chain risk" while another agrees the the same terms that led to that. Either the terms are looser, they're not going to be enforced, or there's another reason for the loud attempt to blacklist Anthropic. It's very difficult to see how you could take this at face value in any case. If it is loose terms or a wink agreement to not check in on enforcement you're never going to be told that. We can imagine other scenerios where the terms stated were not the real reason for the blacklisting, but it's a real struggle (at least for me) to find an explanation for this deal that doesn't paint OpenAI in a very ethically questionable light.
This, for that check theyll be building the autonomous robots themselves, saying "theyre food delivery robots, thats not a gun that a drink dispenser!"
Back in 1960 us early detection systems mistook the moon for a massive nuclear first strike with 99.9% certainty.
With a fully autonomous system the world would have burned.
> Me, and 99% of HN readers, will gladly pull the trigger to release a missile from a drone if we are paid even just US$1,000,000/year.
I sincerely doubt that's true. I hope it's not. $1m is a lot of money, but I find it hard to believe most people would be willing to indiscriminately kill a large number of people for it.
Never mind people in the US, there are plenty of people elsewhere happy to work with their governments who are doubtless developing such autonomous entities.
> Me, and 99% of HN readers, will gladly pull the trigger to release a missile from a drone if we are paid even just US$1,000,000/year.
I will respond with a personal, related story. I was living in Hongkong when "democracy fell" in the late 2010s / early 2020s. It was depressing, and I wanted to leave. (I did later.) I was trying to explain to my parents (and relatives) why most highly skilled foreign workers just didn't care. I said: "Imagine you told a bunch of people in 1984 that they could move to Moscow to open a local office for a wealthy international corporation and get paid big money, like 500K+ in today's dollars. Fat expat package is included. How many people would take it? Most."
Another point completely unrelated to my previous story: Since the advent of pretty good LLMs starting in 2023, when I watch flims with warfare set in the future, it makes absolutely no sense that soldiers are still manually aiming. I'm not saying it will be like Terminator 2 right away, but surely the 19-22 year old operator will just point the weapon in the general direction of the target, then AI will handle the rest. And yet, we still see people manually aiming and shooting in these scenarios. Am I the only one who cringes when I see this? There is something uncanney valley about it, like seeing a character in a film using a flip phone post-2015! Maybe directors don't want to show us the ugly truth of the future of warfare.
I don't cringe because it's for dramatic/narrative effect. It's the same reason the crew of the Enterprise regularly beam into dangerous locations rather than sending a semi-autonomous drone. Or that despite having intelligent machines their operations are often very manual, as it is on many science fiction shows. The audience (if they think about it) realises this is not realistic and understands that the vast majority of our exploration would be done by unmanned/automated vessels. But that wouldn't be very interesting.
Other universes take it further - Warhammer 40k often features combatants fighting with melee weapons. Rule of cool and all that.
Agreed, but I think it goes far beyond warfare. The biggest "plot hole" in much scifi (IMO) is the lack of explanation for why all the depicted systems aren't autonomous. Most worldbuilding seems rather lazy to me, a haphazard mishmash of things that imply AGI and things that would only ever exist in a pre-ChatGPT world.
One of the few works that at least attempts to get this right is the Culture series where it's remarked on several different occasions that anything over some threshold of computing power has AGI built into it (but don't worry you're totally free, just ignore the hall monitor in all of your devices).
I mean this is not actually true and the statement justifies and vindicates those that do sell out by saying of course anyone would. There are countless marytr for religion, politics, and other things.
A better way is to say you can always find a cheap sellout at least than the morally dammed cannot claim equality of belief
> There are countless marytr for religion, politics, and other things.
I think those are not really comparable to OpenAI employees who leave, but that only underlines your point more:
Leaving OpenAI is not like death. In fact most of the employees will have an easy time finding a new job, given the resume of having worked at OpenAI. It is nowhere near any actual martyr.
You mean like all of the religious leaders who are actively supporting a defending a three time married adulterer? You’ll have to excuse my skepticism of the morality of “the moral majority”.
Religion is and always has been about control… it strikes me as exceedingly naive to be surprised the church is backing a pedophile, have you literally ever read any history of any kind?
Not claiming all religious people everywhere are some moral majority? Simply that people die for there beliefs and don't sellout. It happens in religion, politics etc. Also it's some super faulty logic to say look those prominent religious people support trump so all religous people support him is stupid. If that were the case Trump would win every election by massive margins. Trump might win 60/40 in rural areas the 40% he is losing is still very religous generally speaking because rural populations are religious. Cambridge MA voted for Biden by like 96% they have more than 4% of there populations that is also religious.
Also your point is kind of self defeating Trump's true believers dont sell Trump out no matter what he does. He could hide and suppress a pedophile conspiracy and his believers will still say he is tough on crime.
Selling out is bad I think people should passionate stand and be consistent in what they believe and do anything less shouldn't be celebrated or excused because its hard
1) I don't think you have read nor understood my argument. Stating that 80% of Evangelical Christians voted for Trump is not the the ding you think it is. Your imply 1/5 of Evangelicals don't sellout? I think that estimate is way too high and even if it was 99.99% of Evangelic Christians that doesn't excuse their selling hence my original statement. Say everyone sellsout so its okay to sell out is excusing in my opinion is abhorent behavior and suppport of an extremely dangerous leader. But this leads to point 2.
2) I am assuming your a democrat, congrats me to. I am also religiou, and I am assuming your not religous. But you don't see to undertand much about different religous groups and I think this sharp narrow view thinking really harms the democrats ability to reach out to religous people which is around 70-75% of american's according to pew research.
If you want to understand Evangelicals are basically defined by following some charismatic leader who either speaks for Christ, or has visions, or just claims to have all the answers. Believers will follow in any direction because they trust that person but when trust is lost they usually face a crisis of faith and leave that church or the faith all together because they didn't really have strong buy in to the ideals of chirst just that person. This is an extremely well documented occurence. While not all Evangelical people are that occurence a large number are and that architype perfectly describes Trump supporters and MAGA cultist. I think that explains the extreme overlap.
But also religion is a much more complex subject than 1 statistic, as being MAGA is not on the set of beliefs required to be Christian, In fact being a good person isn't either. I think it would be worth while to read up on up and coming people like James Talarico, who understand well that infusing the 2 philosophies is motivating. Because remember religion while being used to do horrible things was also used to immensely liberal things. Universal voting is a protestant thing, anti slavery is largerly a religous movement against white supremacy. The civil rights movement is baked in religion.
Understanding religion as equal to Trump support is corrosive the game of politics and pandering the playroom of vanity. As it doesn't help change anything and is just social meeting points to talk that way. I don't care for vanity I care about wining political power and using it run the country well and help people. I care about uplifting people economically, so they have the freedom to explore whatever faith, athiestism, or whatever they want because that liberality I believe is inherent into the decency of the human condition.
I very much understand religion and I’m surrounded by religious folks as a 52 year old Black guy growing up with religious parents and still living in the Bible Belt and I actually went to a private Christian mostly white school through elementary school.
I understand the difference between socially liberal Christian churches like the ones that were key to the civil rights movement and today are fightijg ICE.
My own wife is what I consider a very liberal Christian. She is a tither and she also is a dance fitness instructor and almost every male fitness I instructor in her organization is gay and she considers them friends and she is as far away from MAGA as possible. (I was a fitness instructor part time for over a decade myself in my younger years and I am well aware that all male instructors aren’t gay).
But the Black led mega churches also aren’t speaking up strongly about all of the things that clearly go up against the “RFC of Christianity” - adultery, bearing false witness, etc.
Yep, theoretically it could just be oligarchic corruption and not institutional insanity at the highest levels of the government. What a reassuring relief it would be to believe that.
I agree with your assessment, but given the past behaviour of this administration I wouldn't be shocked to discover that the real reason is "petulance".
I agree it makes little sense, and I think if all players were rational it never would have played out this way. My understanding is that there are other reasons (i.e., beyond differing red lines) that made the OpenAI deal more palatable, but unfortunately the information shared with me has not been made public so I won't comment on specifics. I know that's unsatisfying, but I hope it serves as some very mild evidence that it's not all a big fat lie.
Your ballooned unvested equity package is preventing you from seeing the difference between “our offering/deal is better” and “designated supply chain risk and threatening all companies who do business with the government to stop using Anthropic or will be similarly dropped” (which is well past what the designation limits). It’s easier being honest.
The supply chain risk stuff is bogus. Anthropic is a great, trustworthy company, and no enemy of America. I genuinely root for Anthropic, because its success benefits consumers and all the charities that Anthropic employees have pledged equity toward.
Whether Anthropic’s clear mistreatment means that all other companies should refrain from doing business with the US government isn’t as clear to me. I can see arguments on both sides and I acknowledge it’s probably impossible to eliminate all possible bias within myself.
One thing I hope we can agree on is that it would be good if the contract (or its relevant portions) is made public so that people can judge for themselves, without having to speculate about who’s being honest and who’s lying.
>Whether Anthropic’s clear mistreatment means that all other companies should refrain from doing business with the US government isn’t as clear to me.
That isn't what many of us are challenging here. We're not concerned about OpenAI's ethics because they agreed to work with the government after Anthropic was mistreated.
We're skeptical because it seems unlikely that those restrictions were such a third rail for the government that Anthropic got sanctioned for asking for them, but then the government immediately turned around and voluntarily gave those same restrictions to OpenAI. It's just tough to believe the government would concede so much ground on this deal so quickly. It's easier to believe that one company was willing to agree to a deal that the other company wasn't.
I’m skeptical because while I can totally believe that the deal presently contains restrictive language, I can totally believe that OpenAI will abandon its ethical principles to create wealth for the people who control it. Sort of like how they used to be a non-profit that was, allegedly, about creating an Open AI, and now they’re sabotaging the entire world’s supply of RAM to discourage competition to their closed, paid model.
Exactly this. Looks like we had the same conclusion. I really am inclined to believe that OpenAI given that its IPO'ing (soon?) would be absolutely decimated and employees would be leaving left and right if they proclaimed that, yes OpenAI is selling DOD autonomous killing machines.
But we all know how OpenAI is desperate for money, its the weakest link in the bubble quite frankly burning Billions and failed at Sora and there isn't much moat as well economically.
DOD giving them billions for a deal feels like a huge carrot on the stick and wink wink (let's have autonomous killing machines) with the skepticism that you, me or perhaps most people of the community would share.
I for what its worth, don't appreciate Anthropic in its whole (I do still remember perhaps the week old thread where everyone pushed on Anthropic for trying to see user data through API when they looked at the chinese models whole thing) but I give credit where its due and Enemy of my Enemy is my friend, and at the moment it seems that OpenAI might be more friendlier to DOD who wishes to create autonomous killing machine and mass surveillance systems which is like Sci-fi level dystopia rather than anthropic.
> One thing I hope we can agree on is that it would be good if the contract (or its relevant portions) is made public
Until they volunteer evidence that the deal is being misdescribed or that it won't be enforced, you can honestly say that you haven't seen any. What a convenient position!
> Whether Anthropic’s clear mistreatment means that all other companies should refrain from doing business with the US government isn’t as clear to me.
You're conflating the Trump administration and their fascist tendencies with all US government. You want to work for fascists if you get paid well enough. You can admit that on here.
Friend, this reads like that situation where your paycheck prevents you from seeing clearly - I forget the exact quote. Sam doesn't play a straight game and neither does the administration - there are more than a few examples.
I agree with what you're saying, but given the egos involved in the current admin there's a practical interpretation:
1. Department of War broadly uses Anthropic for general purposes
2. Minority interests in the Department of War would like to apply it to mass surveillance and/or autonomous weapons
3. Anthropic disagrees and it escalates
4. Anthropic goes public criticizing the whole Department of War
5. Trump sees a political reason to make an example of Anthropic and bans them
6. The entirety of the Department of War now has no AI for anything
7. Department of War makes agreement with another organization
If there was only a minority interest at the department of war to develop mass surveillance / autonomous weapons or it was seen as an unproven use case / unknown value compared to the more proven value from the rest of their organizational use of it, it would make sense that they'd be 1) in practice willing to agree to compromise on this, 2) now unable to do so with Anthropic in specific because of the political kerfuffle.
I imagine they'd rather not compromise, but if none of the AI companies are going to offer them it then there's only so much you can do as a short term strategy.
That is pretty optimistic, i hope it is true, and just a miss-understanding.
But man, this blew up pretty fast for a miss-understanding in some negotiation. Something must have been said in those meetings to make anthropic go public.
These people are drunk on power. They have been running around dictating things to everyone so for someone to push back is pretty novel _and_ it will inspire (I hope) other people to push back.
Nah, they just respectfully said no to their face, which prompted him to make a big threat display and post another message with caps and exclamation signs on social media.
As an OpenAI employee, quitting wouldn't be a problem, as you have a much higher chance of being successful after quitting than anyone else. You could go to any VC and they would fund you.
This isn't even close to true. VCs aren't silly, and it's not the 2010-2015 days of free money any more. Having a big company on your resume is not enough to land your seed round. You need a product, traction, and real money revenue in most cases.
I mean, even if that's the case Facebook was hiring 100 Million$ just a few months ago though even poaching from OpenAI and I do think that these employees will always have an easier time getting a decent job offer from major companies in general as well. They may or may not be making the same money but, I do think that their morals have to be priced in as well.
Yes I agree, I don't know the current VC market so I am not gonna comment about that but my point was that the OpenAI employees would still be considerably well off even if they switch jobs.
My point was I don't think that Money (whether from VC or taking Jobs from other massive AI employers) should be as important issue to them atleast imo.
Yeah, agreed. I probably wasn't going to delete my OpenAI account (ala the link that is also being upvoted on HN), it just seemed like a hassle vs ceasing to use OpenAI. But when the staff at OpenAI employ mental gymnastics, selective hearing, willful ignorance, or plain ignorance to justify compliance with manmade horrors, I think it's probably important to vote with our feet.
> while another agrees the the same terms that led to that
One of them needs to be investigated for corruption in the next few years. I’d have to assume anyone senior at OpenAI is negotiating indemnities for this.
> Respectfully, it's very hard to see how anyone could look at what just happened and come to the conclusion that one company ends up classed a "supply chain risk" while another agrees the the same terms that led to that.
to be clear i think your assessment of this situation is likely, but it could also be the case that pete and co likes sam more than they do dario.
I was trying to make no particular call on the actual reason aside from pointing at how obviously not the real story and false the statements made so far are. What a knot you have to tie yourself into to seek out an explanation where OpenAI has not made an ethical compromise to stay in the game here. I can stretch and think of some ways but they are far from the simplest explanation.
Lots of responses below give the likely real reasons most of which are probably true in part, but my opinion is it's the primary reason all who is in and who is out decisions are made by the trump administration - fealty. Skills, value brought, qualifications, etc. none of that matter above passing frequent loyalty tests, appealing to ego, bribes (sorry, i mean donations). Imagine thinking "hey, we'll work towards fully autonomous killbots because our adversaries will get them too but the tech isn't strong enough to allow them loose yet" or "yes you can use our ai for your panopticon surveillance, but just not on our own citizens because that is illegal" are lefty woke stances but here we are. Dario failed the loyalty test, as anyone rational would.
> one company ends up classed a "supply chain risk" while another agrees the the same terms that led to that
Never discount the possibility of Hegseth being petty and doing the OpenAI deal with the same terms to imply to the world that Anthropic is being unreasonable because another company signed a deal with him.
anthropic has nothing but a contract to enforce what is appropriate usage of their models. there are no safety rails, they disabled their standard safety systems
openai can deploy safety systems of their own making
from the military perspective this is preferable because they just use the tool -- if it works, it works, and if it doesn't, they'll use another one. with the anthropic model the military needs a legal opinion before they can use the tool, or they might misuse it by accident
this is also preferable if you think the government is untrustworthy. an untrustworthy government may not obey the contract, but they will have a hard time subverting safety systems that openai builds or trains into the model
- When has any AI company shipped "safeguards" that aren't trivially bypassed by mid bloggers? Just one example would be fine.
- The conventional wisdom is that OAI's R&D (including safety) is significantly behind Anthropic's.
- OpenAI is constantly starved for funding. They don't make money. They have every incentive to say yes to a deal that entrenches them into govt systems, regardless of the externalities
There's a critical mass of Trump Derangement Syndrome in SV, as this site exemplifies almost daily. The amount of vitriol and hatred spewed here is not healthy, nor are those who spew it. It kills rational debate, nuance and leads to foolish choices like someone cutting off their nose to spite their face as the old saying goes.
The president of the United States sets the tone that hated without reason or explanation is the way the system works now. Belligerence and power are the currency.
Speaking to people's better angels as if it has a chance of influencing Trumps behaviour is a fool's errand. It's not derangement. His word is worthless.
(Disclosure, I'm a former OpenAI employee and current shareholder.)
I have two qualms with this deal.
First, Sam's tweet [0] reads as if this deal does not disallow autonomous weapons, but rather requires "human responsibility" for them. I don't think this is much of an assurance at all - obviously at some level a human must be responsible, but this is vague enough that I worry the responsible human could be very far out of the loop.
Second, Jeremy Lewin's tweet [1] indicates that the definitions of these guardrails are now maintained by DoW, not OpenAI. I'm currently unclear on those definitions and the process for changing them. But I worry that e.g. "mass surveillance" may be defined too narrowly for that limitation to be compatible with democratic values, or that DoW could unilaterally make it that narrow in the future. Evidently Anthropic insisted on defining these limits itself, and that was a sticking point.
Of course, it's possible that OpenAI leadership thoughtfully considered both of these points and that there are reasonable explanations for each of them. That's not clear from anything I've seen so far, but things are moving quickly so that may change in the coming days.
I don't understand how any sort of deal is defensible in the circumstances.
Government: "Anthropic, let us do whatever we want"
Anthropic: "We have some minimal conditions."
Government: "OpenAI, if we blast Anthropic into the sun, what sort of deal can we get?"
OpenAI: "Uh well I guess I should ask for those conditions"
Government: blasts Anthropic into the sun "Sure whatever, those conditions are okay...for now."
By taking the deal with the DoW, OpenAI accepts that they can be treated the same way the government just treated Anthropic. Does it really matter what they've agreed?
It looks like Anthropic likely wanted to be able to verify the terms on their own volition whereas OpenAI was fine with letting the government police themselves.
From the DoD perspective they don't want a situation, like, a target is being tracked, and then the screen goes black because the Anthropic committee decided this is out of bounds.
> From the DoD perspective they don't want a situation, like, a target is being tracked, and then the screen goes black because the Anthropic committee decided this is out of bounds.
Anthropic didn't want a kill switch, they wanted contractual guarantees (the kind you can go to courts for). This administration just doesn't want accountability, that's all.
It was OpenAI that said they prefer to rely on guardrails and less on contracts (the kind that stops the AI from working if you violate). The same OpenAI that was awarded the contract now.
I don’t know why more people don’t see this. It’s a matter of providing strong guarantees of reliability of the product. There is already mass surveillance. There is already life taking without proper oversight.
I think it's a bit more nuance than that. The government (however good or bad, just bear with me) already has oversight mechanisms and already has laws in place to prevent mass surveillance and policy about autonomous killing.
So the governments stance is "We already have laws and procedures in place, we don't want and can't have a CEO to also be part of those checks"
I don't think this outcome would have been any different under a normal blue government either. Definitely with less mud slinging though.
If you think a blue government would even consider threatening to falsely accuse a company of being a supply-chain threat in order to gain leverage in a contract negotiation, you're insane. There's nothing remotely normal about this, it's not something you see in any western democracy
Government's free to not like the terms and go with another provider. That's whatever.
Government's not free to say, "We'll blow up your business with a false accusation if you don't give us the terms we want (and then use defence production act to commandeer the product anyway)". How much more blatantly authoritarian does it get than that?
This is wise analysis. To summarize: appeasement of the Trump administration is a losing strategy. You won’t get what you want and you’ll get dragged down in the process.
Jeremy Lewin's tweet referenced that "all lawful use" is the particular term that seems to be a particular sticking point.
While I don't live in the US, I could imagine the US government arguing that third party doctrine[0] means that aggregation and bulk-analysis of say; phone record metadata is "lawful use" in that it isn't /technically/ unlawful, although it would be unethical.
Another avenue might also be purchasing data from ad brokers for mass-analysis with LLMs which was written about in Byron Tau's Means of Control[1]
The term lawful use is a joke to the current administration when they go after senators for sedition when reminding government employees to not carry out unlawful orders. It’s all so twisted.
To be clear, the sticking point is actually that the DoD signed a deal with Anthropic a few months ago that had an Acceptable Use Policy which, like all policies, is narrower than the absolute outer bounds of statutory limitations.
DoD is now trying to strongarm Anthropic into changing the deal that they already signed!
I’d like to see smart anonymous ways for people to cryptographically prove their claims. Who wants to help find or build such an attestation system?
I’m not accusing the above commenter of deception; I’m merely saying reasonable people are skeptical. There are classic game theory approaches to address cooperation failure modes. We have to use them. Apologies if this seems cryptic; I’m trying to be brief. It if doesn’t make sense just ask.
Did Sam Altman say that he wouldn't allow ChatGPT to be used for fully autonomous weapons? (Not quite the same as "human responsibility for use of force".)
I don't want to overanalyze things but I also noticed his statement didn't say "our agreement specifically says chatgpt will never be used for fully autonomous weapons or domestic mass surveillance." It said something that kind of gestured towards that, but it didn't quite come out and say it. It says "The DoW agrees with these principles, and we put them in our agreement." Could the principles have been outlined in a nonbinding preamble, or been a statement of the DoW's current intentions rather than binding their future behavior? You should be very suspicious when a corporate person says something vague that somewhat implies what you want to hear - if they could have told you explicitly what you wanted to hear, they would have.
But anyway, it doesn't matter. You said you don't think it should be used for autonomous weapons. I'd be willing to bet you 10:1 that you'll never find altman saying anything like "our agreement specifically says chatgpt will never be used for fully autonomous weapons", now or any point in the future.
> you'll never find altman saying anything like "our agreement specifically says chatgpt will never be used for fully autonomous weapons"
To be fair, Anthropic didn't say that either. Merely that autonomous weapons without a HITL aren't currently within Claude's capabilities; it isn't a moral stance so much as a pragmatic one. (The domestic surveillance point, on the other hand, is an ethical stance.)
They specifically said they never agreed to let the DoD use anthropic for fully autonomous weapons. They said "Two such use cases have never been included in our contracts with the Department of War, and we believe they should not be included now: Mass domestic surveillance [...] Fully autonomous weapons"
Their rational was pragmatic. But they specifically said that they didn't agree to let the DoD create fully automatic weapons using their technology. I'll bet 10:1 you won't ever hear Sam Altman say that. He doesn't even imply it today.
Not sure how that's relevant. I never said Dario was taking an ethical stand. I said they did not agree for Claude to be used for fully autonomous weapons. Now, compare that to OpenAI, whose agreement does allow fully autonomous weapons.
> My understanding is that the OpenAI deal disallows domestic mass surveillance and autonomous weapons,
In that case, what on earth just happened?
The government was so intent on amending the Anthropic deal to allow 'all lawful use', at the government's sole discretion, that it is now pretty much trying to destroy Anthropic in retaliation for refusing this. Now, almost immediately, the government has entered into a deal with OpenAI that apparently disallows the two use cases that were the main sticking points for Anthropic.
Do you not see something very, very wrong with this picture?
At the very least, OpenAI is clearly signaling to the government that it can steamroll OpenAI on these issues whenever it wants to. Or do you believe OpenAI will stand firm, even having seen what happened to Anthropic (and immediately moved in to profit from it)?
> and that OpenAI is asking for the same terms for other AI companies (so that we can continue competing on the basis of differing services and not differing scruples)
If OpenAI leadership sincerely wanted this, they just squandered the best chance they could ever have had to make it happen! Actual solidarity with Anthropic could have had a huge impact.
Am I wrong to think that such an agreement is basically meaningless? OpenAI gets to say there are limits, the government gets to do whatever it wants, and OpenAI will be very happy not to know about it.
Bingo. You don’t have to read much into this if you remember how the DoD uses the word trust. In their world, a "trusted" system is one that has the power to break your security if it goes wrong. So when they say "unrestricted use," the likely meaning isn’t just fewer guardrails it’s that the vendor doesn’t get to monitor or audit how the system is being used. In other words, the government isn’t handing a private company visibility into sensitive operations.
"AI shouldn't be used for mass surveillance or autonomous weapons". The statement from OpenAI virtually guarantees that the intention is to use it for mass surveillance and autonomous weapons. If this wasn't the intention them the qualifier "domestic" wouldn't be used, and they would be talking about "human in the loop" control of autonomous weapons, not "human responsibility" which just means there's someone willing to stand up and say, "yep I take responsibility for the autonomous weapon systems actions", which lets be honest is the thinnest of thin safety guarantees.
Assuming this is real: Why do you think anthropic was put on what is essentially an "enemy of the state" list and openai didn't?
The two things anthropic refused to do is mass surveillance and autonomous weapons, so why do _you_ think openai refused and still did not get placed on the exact same list.
It's fine to say "I'm not going to resign. I didn't even sign that letter", but thinking that openai can get away with not developing autonomous weapons or mass surveillance is naive at the very best.
My understanding is that OpenAI's deal, and the deal others are signing, implicitly prevents the use of LLMs for mass domestic surveillance and fully autonomous weapons because today one care argue those aren't legal and the deal is a blanket for allowing all lawful use.
Today it can't be used for mass surveillance, but the executive branch has all the authority it needs to later deem that lawful if it wishes to, the Patriot Act and others see to that.
Anthropic was making the limits contractually explicit, meaning the executive branch could change the line of lawfulness and still couldn't use Anthropic models for mass surveillance. That is where they got into a fight and that is where OpenAI and others can claim today that they still got the same agreement Anthropic wanted.
Life is more than a paycheck. We should raise the bar a little IMO. Turning down money for good reasons is not something extreme we should only expect from saints.
Of course. Doesn't change the reality that this is why someone would accept a justification that a neutral would easily see as plainly dishonest. Anyway, this is why we need unions
Who still does business with open ai and why? They are usually 5th or sixth in the benchmarks bracketed below and above by models that cost less. This has been the case for quite some time. Glm is out for us government purposes I'd imagine, but if google agrees to the same terms I don't see why the us government would use open ai anyway. If google disagrees it would be rather confusing given the other invasions of privacy they have facilitated, but if they do then using open ai would make sense as all that would be left is grok...
Imo the more ethical thing is obstructionism. Twitter's takeover showed it's pretty easy to find True Believer sycophants to hire. Better to play the part while secretly finding ways to sabotage.
Why do you suppose OpenAI's deal led to a contract, while Anthropic's deal (ostensibly containing identical terms) gets it not only booted but declared a supply chain risk?
"domestic" "mass" surveillance, two words that can be stretched so thin they basically invalidate the whole term. Mass surveillance on other countries? Guess that's fine. Surveillance on just a couple of cities that happen to be resisting the regime? Well, it's not _mass_ surveillance, just a couple of cities!
So, can you please draw the line when you will quit?
- If OpenAI deal allows domestic mass surveillance
- If OpenAI allows the development of autonomous weapons
- OpenAI no longer asks for the same terms for other AI companies
Correct?
If so, then if I take your words at face value:
- By your reading non-domestic mass surveillance is fine
- The development of AI based weapons is fine as long as there is one human element in there, even if it could be disabled and then the weapon would work without humans involved
- The day that OpenAI asks for the same terms for other AI companies and if those terms are not granted then that's also fine, because after all, they did ask.
I have become extremely skeptical when seeing people whose livelihood depends on a particular legal entity come out with precise wording around what does and does not constitute their red line but I find it fascinating nonetheless so if you could humor me and clarify I'd be most obliged.
Thank you for responding. Everyone wants to think they will “do the right thing” when their own personal Rubicon is challenged. In practice, so many factors are at play, not least of which are the other people you may be responsible for. The calculus of balancing those differing imperatives is only straightforward for those that have never faced this squarely. I’ve been marched out of jobs twice for standing up for what I believed to be right at the time. Am still literally blacklisted (much to the surprise of various recruiters) at a major bank here 8 years after the fact. I can’t imagine that the threat of being blacklisted from a whole raft of companies contracting with a known vindictive regime would make the decision easier.
The founders are all on a first name basis. I’m surprised no one has noted that Anthropic and OpenAI winning together by giving the world two different choices, just like the US does in its political landscape. In this circumstance, OpenAI wins the local market for its government and aligned entities (while having the free consumer by a matter of cost dynamic for that ideal customer profile which is vary broad and similar to Google’s search audience where most their revenue still depends), while Anthropic is provided the global market and prosumer market where people can afford choice by paying for it.
You should quit because the only reasonable thing for your leadership to have done is to refuse to sign any agreement with DoW whatsoever while it's attempting to strongarm Anthropic in this fashion.
It doesn't even matter if OpenAI is offered the same terms that Anthropic refused. It's absurd to accept them and do business with the Pentagon in that situation.
If you take the government at its word, it's killing Anthropic because Anthropic wanted to assert the ability to draw _some_ sort of redline. If OpenAI's position is "well sucks to be them", there's nothing stopping Hegseth from doing the same to OpenAI.
It doesn't matter at all if OpenAI gets the deal at the same redline Anthropic was trying to assert. If at the end of this the government has succeeded in cutting Anthropic off from the economy, what's next for OpenAI? What happens next time when OpenAI tries to assert some sort of redline?
What's the point of any talk of "AI Safety" if you sign on to a regime where Hegseth (of all people) can just demand the keys and you hand them right over?
#1 weekend HN is not a sane place. #2 emotions are high. #3 for what it’s worth @tedsanders I understand where you’re coming from and I believe you’re making the right choice by staying or at least waiting to make a decision. Don’t let #1 and #2 hurt you emotionally or force you to make a rash decision you later regret.
Edit: I don’t work at OpenAI or in any AI business and my neck is on the chopping block if AI succeeds… like a lot of us. Don’t vilify this guy trying to do what’s right for him given the information he has.
> My understanding is that the OpenAI deal disallows domestic mass surveillance and autonomous weapons
And you believe the US government, let alone the current one will respect that? Why? Is it naïveté or do you support the current regime?
> If it turns out that the deal is being misdescribed or that it won't be enforced, I can see why I should quit.
So your logic is your company is selling harmful technology to a bunch of known liars who are threatening to invade democratic countries, but because they haven’t lied yet in this case (for lack of opportunity), you’ll wait until the harm is done and then maybe quit?
I’ll go out on a limb and say you won’t. You seem to be trying really hard to justify to yourself what’s happening so you can sleep at night.
Know that when things go wrong (not if, when), the blood will be on your hands too.
His point reeks of cope. But making a large amount of money would make anyone dumb, deaf, and blind. Also, I give a little leeway to people who are employees without executive decision-making power, as they do stand to have a lot to lose in situations like this.
It's probably how they are coping with the cognitive dissonance. I certainly feel for them, I don't know that I could easily walk away from a big pay package either without backup options when I have family to support and I'm not near retirement.
Ted, what do you think of your CEO’s statement: “the DoW displayed a deep respect for safety and a desire to partner to achieve the best possible outcome.”
The evidence seems to overwhelmingly point in the opposite direction.
This seems like the kind of foolishness it takes a lot of money to believe. Anthropic blew up their contract with the Pentagon over concerns on lethal autonomous weapons and mass domestic surveillance. OpenAI rushes in to do what Anthropic wouldn't.
If you think that means your company isn't going to be involved in lethal autonomous weapons and mass domestic surveillance... I don't really know what to tell you. I doubt you really believe that. Obviously you will be involved in that and you are effectively working on those projects now.
Aside from that unlikely read, this deal was still used as a pressure point on Anthropic, there's absolutely no way OpenAI was not used as a stick to hit with during negotiations.
Anthropic is deemed a betrayer and a supply chain risk for actually enforcing their principles.
OpenAI agrees to be put in the same position as Anthropic.
It seems like you must actually somehow believe that history will repeat itself, Hegseth will deem OpenAI a supply chain risk too, then move to Grok or something?
There's surely no way that's actually what you believe...
Giving you the benefit of the doubt and assuming [1] does not play a role in your thinking:
I don't mean this in any way rude and I apologize if this comes accross as such but believing it won't be used in exactly this way is just naive. History has taught us this lesson again and again and again.
What people don't understand is that domestic surveillance by the government doesn't happen and isn't needed. They know it's illegal and unpopular and for over two decades they have a loophole. Since the Bush administration it's been arranged for private contractors to do the domestic surveillance on the government's behalf. Entire industries have been built around creating "business records" for no other purpose than to sell them to the government to support domestic surveillance. This is entirely legal and why the DoW has been able to get away with saying things like "domestic surveillance is illegal, we don't do that" for over two decades while simultaneously throwing a shit fit about needing "all legal uses" if their access to domestic surveillance is threatened.
There's a big difference between "the government won't use our tools for domestic surveillance" (DoW/DoD/OpenAI/etc) and "we won't allow anyone to use our tools to support domestic surveillance by the government" (Anthropic)
Hegseth and the current Trump admin are completely incompetent in execution of just about everything but competent administrations (of both parties) have been playing this game for a long time and it's already a lost cause.
Listen, if the Government using it for legit and safe use cases wasn’t an issue, then they wouldn’t have complained about Anthropic’s language. Sam is just looking the other way and pretending for you employees.
Or Sam bribed the government to do this, which is also entirely possible.
I don't know you, so maybe you're actually for real and speaking on good faith here but honestly this and your other responses in this thread read exactly like "...salary depends on not understanding"
Assuming this isn't a troll and you really think this, you should at least have the cojones to admit you're taking the blood money instead of trying to pretzel the truth so hard that you just look like a moron instead.
For the record I don’t care if you quit or not. Cash rules after all… However, you are incredibly naive if you think the current admin will follow through on those terms.
I have a bridge to Brooklyn to sell you if you believe this.
Standing up for whats right often is not easy and involves hard choices and consequences, your leader has shown you and the world that he is not to be trusted.
I can't tell you what to do but I hope you make the right decision.
Looks to me like you have decided that you are being paid to shut up and take the word of the most thoroughly dishonest and corrupt US government we've yet seen. Why on God's slowly-browning green earth do you trust that Altman got the deal Anthropic was trying for?
lol, naive as hell. why would your company's agreement be the same as the one who just refused the _same_ agreement? Even my question doesn't even make sense, this is a contradiction, therefore your statement must be false. There, it's proven
Can you at least stop lying to yourself? Given what they did with Anthropic for not supporting domestic mass surveillance and autonomous weapons...
> My understanding is that the OpenAI deal disallows domestic mass surveillance and autonomous weapons
Your understanding is entirely wrong. At least stop lying to yourself and admit that you are entirely fine with working on evil things if you are paid enough.
I know the money is good, but if I were you (or any OpenAI employee), I'd move over to Google or Anthropic posthaste.
Is it really worth the long-term risk being associated with Sam Altman when the other firms would willingly take you and probably give you a pay bump to boot?
It doesn't make sense to me why anyone would want to associate themselves with Altman. He is universally distrusted. No one believes anything he says. It's insane to work with a person who PG, Ilya, Murati, Musk have all designated a liar and just general creep.
Defending him or the firms actions instantly makes you look terrible, like you'll gladly take the "Elites vs UBI recipients" his vision propagates.
You work for a company that’s part of the Trump, Ellison, Kutchner orbit of corruption.
Y’all are developing amazing technology. But accept reality and drop whatever sense of moral righteousness you’re carrying here. Not because some asshole on the internet says so, but for your own mental health.
there is a recent post about how one of the top OpenAI exec has given 25 million$ to a Trump PAC before publicly supporting Anthropic/signing this deal.
One got characterized as supply chain risk and so much for OpenAI to get the same.
And even that being said, I can be wrong but if I remember, OpenAI and every other company had basically accepted all uses and it was only Anthropic which said no to these two demands.
And I think that this whole scenario became public because Anthropic denied, I do think that the deal could've been done sneakily if Anthropic wanted.
So now OpenAI taking the deal doesn't help with the fact that to me, it looks like they can always walk back and all the optics are horrendous to me for OpenAI so I am curious what you think.
The thing which I am thinking OTOH is why would OpenAI come and say, hey guys yea we are gonna feed autonomous killing machines. Of course they are gonna try to keep it a secret right before their IPO and you are an employee and you mention walking out of openAI but with the current optics, it seems that you/other employees of OpenAI are also more willing to work because evidence isn't out here but to me, as others have pointed out, it looks like slowly boiling the water.
OpenAI gets to have the cake and eat it too but I don't think that there's free lunch. I simply don't understand why DOD would make such a high mess about Anthropic terms being outrageous and then sign the same deal with same terms with OpenAI unless there's a catch. Only time will tell though how wrong or right I am though.
If I may ask, how transparent is OpenAI from an employees perspective? Just out of curiosity but will you as a employee get informed of if OpenAI's top leadership (Sam?) decided that the deal gets changed and DOD gets to have Autonomous killing machine. Would you as an employee or us as the general public get information about it if the deal is done through secret back doors. Snowden did show that a lot of secret court deals were made not available to public until he whistleblowed but not all things get whistleblowed though, so I am genuinely curious to hear your thoughts.
Your response is a perfect encapsulation of "It is difficult to get a man to understand something when his salary depends upon his not understanding it."
I think its wrong for someone to ask someone to resign but acting that there is no issue here is debating in bad faith.
The comment perfectly exemplifies the kind of person that would work at OpenAI. Government AI drones could be executing citizens in the streets but they’d still find some sort of cope why it’s not a problem. They’ll keep moving the goalposts as long as the money keeps coming.
One consequence of creating a country of geniuses in a data center is that you now have a country of geniuses who can potentially help your competitors catch up on research, coding, and data labeling. It's a tough problem for the industry and, more importantly, for long-term safety.
We're obviously nowhere close now, but if we get to a world AI becomes powerful, and powerful AI can be used to create misaligned powerful AI, you may have to start regulating powerful AI like refined uranium processing tech, which is regulated more heavily than refined uranium itself.
The people who would otherwise be affected by spam calls, spam messages, ransomware / computer viruses, fake / deceptive websites, or bioengineered viruses.
The risk of these could plausibly increase in a world with powerful AI. Obviously the risk isn't high now, and there are benefits to trade off against these costs, but all powerful technologies have costs.
Bribing employees to disclose confidential information entrusted to them is not kosher nor wholesome. I consider corporate insider trading on these markets to be analogous - if you're an employee and you trade, you are selling your employer's info for money. Nearly every employer would fire employees caught giving away confidential information for personal bribes.
In the stock market, Matt Levine likes to say that insider training is about theft, not fairness. You can be prosecuted for merely sharing info with a friend on a golf course who then proceeds to trade. Your crime is not trading (you didn't even trade), but misappropriating information you were entrusted with and not authorized to sell.
The market economy is not about fairness but about ruthless power.
The worlds most influential people demonstrate that only power matters; that the world order we built last century through unimaginable suffering and violence matters less than securing their own personal gain; that law, morals, and order were just dreams of the weak
- One thing to be aware of is that LLMs can be much smarter than their ability to articulate that intelligence in words. For example, GPT-3.5 Turbo was beastly at chess (1800 elo?) when prompted to complete PGN transcripts, but if you asked it questions in chat, its knowledge was abysmal. LLMs don't generalize as well as humans, and sometimes they can have the ability to do tasks without the ability to articulate things that feel essential to the tasks (like answering whether the bicycle is facing left or right).
- Secondly, what has made AI labs so bullish on future progress over the past few years is that they see how little work it takes to get their results. Often, if an LLM sucks at something that's because no one worked on it (not always, of course). If you directly train a skill, you can see giant leaps in ability with fairly small effort. Big leaps in SVG creation could be coming from relatively small targeted efforts, where none existed before.
We’re literally at the point where trillions of dollars have been invested in these things and the surrounding harnesses and architecture, and they still can’t do economically useful work on their own. You’re way too bullish here.
Amodei isn't a grifter; the difference is that he really believes powerful AI is imminent.
If you truly believe powerful AI is imminent, then it makes perfect sense to be worried about alignment failures. If a powerless 5 year old human mewls they're going to kill someone, we don't go ballistic because we know they have many years to grow up. But if a powerless 5 year old alien says they're going to kill someone, and in one year they'll be a powerful demigod, then it's quite logical to be extremely concerned about the currently harmless thoughts, because soon they could be quite harmful.
I myself don't think powerful AI is 1-2 years away, but I do take Amodei and others as genuine, and I think what they're saying does make logical sense if you believe powerful AI is imminent.
how long has he believed i? i only watched the first couple of minutes of the interview before coming to my senses, but something something about not having changed his outlook since 2017.
maybe if he can really (but really really) keep believing for 10 more years, we can have this discussion again around that time.
We had access to the eval data (since we funded it), but we didn't train on the data or otherwise cheat. We didn't even look at the eval results until after the model had been trained and selected.
If you don't believe me, that's fair enough. Some pieces of evidence that might update you or others:
- a member of the team who worked with this eval has left OpenAI and now works at a competitor; if we cheated, he would have every incentive to whistleblow
- cheating on evals is fairly easy to catch and risks destroying employee morale, customer trust, and investor appetite; even if you're evil, the cost-benefit doesn't really pencil out to cheat on a niche math eval
- Epoch made a private held-out set (albeit with a different difficulty); OpenAI performance on that set doesn't suggest any cheating/overfitting
- Gemini and Claude have since achieved similar scores, suggesting that scoring ~40% is not evidence of cheating with the private set
- The vast majority of evals are open-source (e.g., SWE-bench Pro Public), and OpenAI along with everyone else has access to their problems and the opportunity to cheat, so FrontierMath isn't even unique in that respect
For our teams, Codex is a massive productivity booster that actually increases the value of each dev. If you check our hiring page, you’ll see we are still hiring aggressively. Our ambitions are bigger than our current workforce, and we continue to pay top dollar for talented devs who want to join us in transforming how silicon chips provide value to humans.
Akin to how compilers reduced the demand for assembly but increased the demand for software engineering, I see Codex reducing the demand for hand-typed code but increasing the demand for software engineering. Codex can read and write code faster than you or me, but it still lacks a lot of intelligence and wisdom and context to do whole jobs autonomously.
reply