This is ignoring the fact that the government is the foundation of society (I know some will disagree with that, but the end result is just government with more steps).
Private models in a low trust society means the government will come and seize the models. Competitive business will only be allowed through cronyism.
The better option is to opt for high trust. Yes the Gman can rip your servers apart, but they know they'll face consequences, legal and political. Laws and regulations are the answer, not locking down into smaller fiefdoms.
You get high trust through social norms, not by more "laws and regulations". Social norms can't be imposed by fiat, they arise spontaneously, often for unclear reasons. That's why they're so fragile and precious. With Trump's destruction of social norms around the presidency and the federal government generally, the US is now just another country where bribery is the cost of doing business.
Also this falls into the "right to bear arms" thing: if LLMs are limited legally, then illegal LLMs will be the superior choice. This is pretty much the plot of Cryptonomicon and Corey's take on I, Robot
Except there’s a large hardware barrier to entry, which for now seems effective.
Related note. Cryptography has been subject to export controls for years and manufacturers bend into pretzels to meet the laws, regulations, and policies.
It's honestly not the worst strategy: make the dangerous move when you have the most tolerance, and then everyone can figure stuff out and make the landing on Sunday.
Options and futures don’t wait and a lot of stuff trades 24x7. You can do your puts right now, and banks and market makers will meet you now if you’re big enough. The landing for Main Street will be more of a horrible traffic accident that happened days ago and they just woke up in the flaming wreck of their financial life.
I personally don't have strong experience here , but I would treat them similar to BUILD files and the like - probably in the root directory of a repo but nowhere near the bin/ or build/ directories.
Also it looks like there's a compilation step to these files, which is interesting. The raw file was included, not the environment specific file.
A bit more complicated, as the aircraft itself was unable to detect the stall conditions due to icing of the pitot tubes so the warning itself was in and out several times. Clearly the copilots did not understand the situation so an inconsistent alarm could be seen as spurious or a secondary effect.
> At the same time he made an abrupt nose-up input on the side-stick, an action that was unnecessary and excessive under the circumstances. The aircraft's stall warning sounded briefly twice due to the angle of attack tolerance being exceeded
...
> The crew's lack of response to the stall warning, whether due to a failure to identify the aural warning, to the transience of the stall warnings that could have been considered spurious, to the absence of any visual information that could confirm that the aircraft was approaching stall after losing the characteristic speeds, to confusing stall-related buffet for overspeed-related buffet, to the indications by the flight director that might have confirmed the crew's mistaken view of their actions, or to difficulty in identifying and understanding the implications of the switch to alternate law, which does not protect the angle of attack.
Its a complicated interplay of systems, where autonomous control systems are changing modes and receiving bad information during a complex, raplidly developing situation.
>A bit more complicated, as the aircraft itself was unable to detect the stall conditions due to icing of the pitot tubes so the warning itself was in and out several times.
74 times the stall warning blared [1]
Of the 3 pilots in the cockpit, only one thought he had to pull up, see page 31, unfortunately he was one of the ones in control.
>raplidly developing situation.
It was the same situation from when it began to the end, stuck pitot tubes. Though the stall warning only started blaring when the pilot stalled the plane. Bad airspeed indicators don't stall the plane, and are something pilots are supposed to be able to handle, that's why 2 of the 3 were shocked one did the exact opposite in the situation.
It was pilot error. Just look at the report, every finding starts with "the Crew". Planes aren't supposed to crash into the ground just because an air speed sensor failed.
Its clearly propaganda. "Your data belongs to you." I'm sure the ToS says otherwise, as OpenAI likely owns and utilizes this data. Yes, they say they are working on end-to-end encryption (whatever that means when they control one end), but that is just a proposal at this point.
Also their framing of the NYT intent makes me strongly distrust anything they say. Sit down with a third party interviewer who asks challenging questions, and I'll pay attention.
"Your data belongs to you" but we can take any of your data we can find and use it for free for ever, without crediting you, notifying you, or giving you any way of having it removed.
…”as does any culpability for poisoning yourself, suicide, and anything else we clearly enabled but don’t want to be blamed for!”
Edit: honestly I’m surprised I left out the bit where they just indiscriminately scraped everything they could online to train these models. The stones to go “your data belongs to you” as they clearly feel entitled to our data is unbelievably absurd
>…”as does any culpability for poisoning yourself, suicide, and anything else we clearly enabled but don’t want to be blamed for!”
Should walmart be "culpable" for selling rope that someone hanged themselves with? Should google be "culpable" for returning results about how to commit suicide?
There are current litigation efforts to hold Amazon liable for suicides committed by, in particular, self-poisoning with high-purity sodium nitrite, which, in low concentrations is used as a meat curing agent.
A 2023 lawsuit against Amazon for suicides with sodium nitrite was dismissed but other similar lawsuits continue. The judge held that Amazon, “… had no duty to provide additional warnings, which in this case would not have prevented the deaths, and that Washington law preempted the negligence claims.“
This is as unproductive as "guns don't kill people, people do." You're stripping all legitimacy and nuance from the conversation with an overly simplistic response.
What? The claim is true. The nuance is us discussing if it should be true/allowed. You're simplifying the moral discussion and overall just being rude/dismissive.
Comparing rope and an LLM comes across as disingenuous. I struggle to believe that you believe the two are comparable when it comes to the ethics of companies and their impact on society.
> Comparing rope and an LLM comes across as disingenuous.
What makes you feel that? Both are tools, both have a wide array of good and bad uses. Maybe it'd be clearer if you explained why you think the two are incomparable except in cases of disingenuousness?
Remember that things are only compared when they are different -- you wouldn't often compare a thing to itself. So, differences don't inherently make things incomparable.
> I struggle to believe that you believe the two are comparable when it comes to the ethics of companies and their impact on society.
I encourage you to broaden your perspectives. For example: I don't struggle to believe that you disagree with the analogy, because smart people disagree with things all the time.
What kind of a conversation would such a rude, dismissive judgement make, anyways? "I have judged that nobody actually believes anything that disagrees with me, therefore my opinions are unanimous and unrivaled!"
A rope isn’t going to tell you to make sure you don’t leave it out on your bed so your loved ones can’t stop you from carrying out the suicide it helped talk you in to.
You are 100% right, a rope likely isn't going to tell you anything. There's one of those differences I mentioned which makes comparisons useful. We could probably name a few differences!
So, what makes you think comparing the 2 tools is invalid? You just compared them yourself, and I don't think you were being disingenuous.
Just because I used italics to emphasize something one time doesn’t mean you get to talk to me like that. I am not a child and you’re being unnecessarily patronizing.
I let it slide in the previous comment and gave you the benefit of the doubt despite what I saw but this comment clearly illustrates how disrespectful you’re being.
I think you, as you put it, rudely, patronizingly, disrespectfully responded to the wrong post: mine was a polite one about a comparison between 2 tools and your statement that the comparing posters must be acting in bad faith (whereas you, with your differing opinion, are acting in good faith).
I'm not interested in focusing on tone-policing, since it is one of the lowest forms of debate and usually avoids the substance of the matter. So, I'm happy to return to our discussion about the 2 tools anytime you want to review my previous post and respond to the substance of it. If you're not into that, have a nice day comfortable in the knowledge that I've already turned the other cheek.
The same that happens with chatgpt? ie. if you do it in an overt way you get a canned suicide prevention result, but you can still get the "real" results if you try hard enough to work around the safety measures.
The moment we learned ChatGPT helped a teen figure out not just how to take their own life but how to make sure no one can stop them mid-act, we should've been mortified and had a discussion.
But we also decided via Sandy Hook that children can be slaughtered on the altar of the second amendment without any introspection, so I mean...were we ever seriously going to have that discussion?
my point is, clearly there is a sense of liability/responsibility/whatever you want to call it. not really the same as selling rope, rope doesn't come with suicide warnings
Throw on taxes, administrative overheads, etc, they are probably looking at 30-45 sales per month. Which is likely not realistic.
On top of that, this is a continuous payment. Even if I was looking at 5-10x rate of return, I would be very hesitant as that's the rate-of-return today while the sales are forever.
I've been wondering how realistic microsubscriptions are... Say $1-2 dollars a month per user to maintain an app, perhaps limited to just power users, would support a lot of infrastructure.
Your understanding of stuxnet is flawed, Iran was attacked by the Us Gov in a very very specific spearfish attack with years of preparation to get Stux into the enrichment facilities - nothing to do with lifts connected to the network.
Also the facility was air-gapped, so it wasn't connected to ANY outside network. They had to use other means to get Stux on those computers and then used something like 7 zero days to move from windows into Siemens computers to inflict damage.
Stux got out potentially because someone brought their laptop to work, the malware got into said laptop and moved outside the airgap from a different network.
"Stux got out potentially because someone brought their laptop to work, the malware got into said laptop and moved outside the airgap from a different network."
The lesson here is that even in an air-gapped system the infrastructure should be as proprietary as is possible. If, by design, domestic Windows PCs or USB thumb drives could not interface with any part of the air-gapped system because (a) both hardwares were incompatible at say OSI levels 1, 2 & 3; and (b) software was in every aspect incompatible with respect to their APIs then it wouldn't really matter if by some surreptitious means these commonly-used products entered the plant. Essentially, it would be almost impossible† to get the Trojan onto the plant's hardware.
That said, that requires a lot of extra work. By excluding subsystems and components that are readily available in the external/commercial world means a considerable amount of extra design overhead which would both slow down a project's completion and substantially increase its cost.
What I'm saying is obvious, and no doubt noted by those who've similar intentions to the Iranians. I'd also suggest that the use of individual controllers etc. such as the Siemens ones used by Iran either wouldn't be used or they'd need to be modified from standard both in hardware and with the firmware (hardware mods would further bootstrap protection if an infiltrator knew the firmware had been altered and found a means of restoring the default factory version).
Unfortunately, what Stuxnet has done is to provide an excellent blueprint of how to make enrichment (or any other such) plants (chemical, biological, etc.) essentially impenetrable.
† Of course, that doesn't stop or preclude an insider/spy bypassing such protections. Building in tamper resistance and detection to counter this threat would also add another layer of cost and increase the time needed to get the plant up and running. That of itself could act as a deterrent, but I'd add that in war that doesn't account for much, take Bletchley and Manhattan where money was no object.
I once engineered a highly secure system that used (shielded) audio cables and amodem as the sole pathway to bridge the airgap. Obscure enough for ya?
Transmitted data was hashed on either side, and manually compared. Except for very rare binary updates, the data in/out mostly consisted of text chunks that were small enough to sanity-check by hand inside the gapped environment.
Stux also taught other government actors what's possible with a few zero days strung together, effectively starting the cyberwasr we've been in for years.
Private models in a low trust society means the government will come and seize the models. Competitive business will only be allowed through cronyism.
The better option is to opt for high trust. Yes the Gman can rip your servers apart, but they know they'll face consequences, legal and political. Laws and regulations are the answer, not locking down into smaller fiefdoms.
reply