My father used to have pain in his abdomen out of nowhere with a couple years in between, but they never could figure out what was wrong with him even during an extended hospital stay at one point. I think the earliest time I must have been in elementary school, but my first semester in college during the weekend where my parents were going to come down to visit for the first time, he had the recurring pain again that got so much worse than usual that he was jaundiced and my mother forced him to go to the hospital again (medical situations are basically the only circumstance where I've seen him get irritable), and it turned out that his gall bladder was completely full of stones that they said would have started spilling over into his liver and put that at risk as well if he delayed coming in for a few more hours. They removed his gall bladder, and he's never felt any symptoms in the decade since.
> Logically extrapolated, in this scenario, the airline becomes responsible for the baby's wellbeing. And I don't think that's a good place for the airlines to be?
Aren't airlines responsible for the well-being of everyone on a plane? That seems like it kind of goes with the territory of charging money for hurtling people across the sky in a giant metal box.
I think this is ill-phrased, but true. It isn't about being responsible for wellbeing, it is about legal liability in case of accidents.
If you (as the airline) did what the rules where that everybody else follows and the FAA mandates and the baby dies you can argue you did everything correctly and the rules are at fault.
If you however go your own special, non-FAA-approved solution, typically the burden of proof that this special solution is at least up to par with existing certified standards is on you, the airline and if a baby dies it is on you alone to defend that solution.
Meaning from the standpoint of an airline there is not a lot to gain, but much to lose if they do things outside established standards.
This may sound idiotic, but it also prevents other "creative" solutions where airlines sacrifice safety willingly to spend less money.
I don't disagree with that. To me, that implies that there should be better established standards. The parent comment described some of the mechanisms, but stopped short of suggesting that they become standard, which I think is why I was sort of confused.
>This may sound idiotic, but it also prevents other "creative" solutions where airlines sacrifice safety willingly to spend less money.
Everyone loves to say this from the comfort of their arm chair somewhere in an ivory tower but literally every aviation and adjacent workplace has a phrase to the tune of "here we cut stupid corners, cutting smart ones is illegal"
Which is to say, the do-gooders in their infinite wisdom have blocked progress on various axis, some of them no-brainers, some of them the most reasonable ones, because once upon a time someone took it too far and something happened. So all investigation of where to save money, where to find improvement, etc, etc, happens on axis of optimization that are frequently absurd. This is where most of the "creative" solutions you deride come from.
> because once upon a time someone took it too far and something happened.
It's easier to accept red tape once you follow "something happened" down to the idiots who "took it too far". Progress is blocked along the axes your fuckups have access to, end of story.
If you don't like red tape quit playing grab-ass around heavy machinery.
>It's easier to accept red tape once you follow "something happened" down to the idiots who "took it too far". Progress is blocked along the axes your fuckups have access to, end of story. If you don't like red tape quit playing grab-ass around heavy machinery.
<sigh> This is exactly the sort of sneering smugness that makes people rightly hate the safety guy and you are a bad person for as long as you continue to engage in it.
> These two statements are at odds. You first assuming that lack of trust would be the only reason he needs to know what happens next, then go on to outline another reason he and the larger organization would both benefit from knowing more.
But doesn't that same logic apply to whether they need to know the details?
No. Leaderships job in this context is to set expectations of what happens next and to drive the team towards a solution that meets those expectations. No amount of trust magically makes that expectation materialize or for it to be met.
Leadership doesn't need to know what happened to set this expectation and then trust their team - who already know the details - to figure out how to meet what's expected.
Perhaps my favorite quote ever about leadership is that it's about holding a vision long enough for others to realize it for themselves, but you can't do that if you dictate and micromanage.
So if I'm understanding you correctly, you're saying there's literally no reason for a leader to need to know the details ever unless they don't trust the people implementing the plan. I don't agree with that at all.
Leaders have the option of trusting subordinates, then operating at a higher level. However part of the job of a leader is to decide who to trust, with what. And why.
Sometimes that means, "Trust, but verify." The leader will spot check randomly to see that the whole is good.
Often, as here, that means, "Trust in some things, but not others." The leader in this case trusted the technical person to produce an accurate description of what happened. But didn't trust that same person to arrive at a solution that met the business of having a similar disaster not happen the next time.
But rather than undermine the tech person with, "I don't trust you to meet the business need," the leader communicated it with, "I trust you on the technical details, but here is the expectation for what you need to produce."
The statement I initially responded to was "You first assuming that lack of trust would be the only reason he needs to know what happens next, then go on to outline another reason he and the larger organization would both benefit from knowing more". I pointed out that the same logic seems like it would apply to the idea that there might be other reasons why a leader might need to know the details. They said "no", and then a bunch of other stuff that you seem to think that I'm not reading. I'm reading it, and I'm reading what you said to add onto their comment, I just don't understand how you can claim their answer of "no" to my question doesn't mean "no" to the restatement of my question.
Leaders should not need to know every single detail (ie how this incident happens, or all the tiny process changes that are happening toward their visions). That is different than knowing nothing in detailed.
The baseline expectation in the original post also applies to the SVP: he knew enough to know that things aren't unreasonable to start with, this expectation might not apply to all companies.
An analogy that might make more sense is horizontal scalability vs. vertical scalability. A leader's ability to understand all the details of the proposal brought to them by the team inherently caps the organization's ability to scale up.
Offloading the detail onto the team allows the org to scale higher by allowing the leader to bring more teams/problems into the mix. But it also introduces the complexity of keeping those teams in close enough alignment that the overall effect is still positive.
This is by no means guaranteed to be positive, and there's a lot of details to get right for leadership in making this work. But they are different details than what each individual team was bringing to leadership.
Abstractions are helpful, in business management as much as in systems architecture.
The top-level comment that started this discussion was mentioning that a "leader" said "I don't need to know the details" but that they wanted to know what comes next. The comment I responded to claimed that there are valid reasons to need to know what comes next that don't stem from a lack of trust. I asked whether the same logic should apply to wanting to know the details (since to me it seems pretty reasonable that it should). The response said "no".
It seems like you're trying to rebut my argument by presenting my argument as being against a less strong version of the original claim rather than understanding it as being literally just a counterargument to the specific version of the claim I thought was too strong.
You, and this entire thread is missing the point. Nobody is wrong but, the point being missed is that "leaders" isn't one thing.. there are leaders at every height of the organizations hierarchy and even to the sides. As a basic example, do you mean the immediate manager? their manager? their manager? The CTO? Each operates at ever more abstract level of detail.
What do you think knowing the details will do for the person in the leadership position? It doesn't change what needs to be done, it doesn't change what has been done, all it usually does it put a monkey on the leaders back that's best kept with the team that created it.
Worse it invites bike shedding most of the time, where folks sit in a room nitpicking pointless details that's cathartic for the people in charge, but once again, doesn't actually do anything productive.
> The statement I initially responded to was "You first assuming that lack of trust would be the only reason he needs to know what happens next, then go on to outline another reason he and the larger organization would both benefit from knowing more". I pointed out that the same logic seems like it would apply to the idea that there might be other reasons why a leader might need to know the details.
I'm genuinely confused why this is apparently not clear. Someone said that it's flawed logic to assume there aren't good reasons beyond a lack of trust to want to know next steps. I'm saying it seems reasonable there could be good reasons beyond a lack of trust to want to know the details of the current step. You're saying that I'm finding it difficult to "grok this concept", but I don't even know what concept you're talking about, because none of the responses seem to have been related to this, and I don't understand what anyone else could think I've been saying.
I can understand honest mistakes, but like, usually when I develop anything with agents (which I assume is what you're doing internally at Anthropic), they're almost too enthusiastic about trying to add test cases to the point where they sometimes try to glue together things in ways that are structurally impossible in the actual code in order to try to test that behavior. I'm honestly a bit mystified that adding a new feature didn't get bundled in with tests that the feature works for arbitrary configurations.
My first thought was the irony of something used to interface with models that were trained on data that was presumably not properly licensed having the same name as the Creative Commons abbreviation.
This is super interesting to me. I've slowly been working on something similar (https://gitlab.com/saghm/tartarus) because my ideal sandboxing is "prevent writing to anything outside this dir but still allow reading to most things so that I don't have to manually copy things into a container/VM". I approached it by trying to figure out how to build up a bubblewrap based on a config that gave the properties I wanted, with the hope that I could eventually expand it to support other platforms via stuff like `sandbox-exec` on MacOS, but I haven't had time to work on it more for a while.
At a glance, this seems to be providing most of what I was originally looking for when I ended up deciding I'd have to write it myself, but focusing specifically on Linux and providing a more full-fledged sandbox rather than only caring about a small set of permissions that I personally had a need for. Probably the biggest (and least hardened) feature that I spent time on in mine was trying to figure out how to allow arbitrary GUI apps so that I could run agents in it via Zed.
I'm definitely going to try this out and see how well it works for me. It's insane to me that this is something none of the big AI companies have bothered solving this yet other than via opaque rules built into their harnesses or absolutely awful manual rules that expect me to hard-code shapes of shell commands that I want to allow or not allow.
> my ideal sandboxing is "prevent writing to anything outside this dir but still allow reading to most things so that I don't have to manually copy things into a container/VM"
That's what Codex does out of the box, and it's not good against malware - i.e. a rogue npm packet (or even just codex after prompt injection) can read your ssh key and send it to the attacker.
As I said, opaque rules built into the harness rub me the the wrong way. They could change in an update without anything making it clear. Plus, I don't use Codex outside of work (I don't have any active paid subscriptions LLM offerings).
> it's not good against malware - i.e. a rogue npm packet (or even just codex after prompt injection) can read your ssh key and send it to the attacker
Ignoring the repeated references to software I don't personally use, I never said I was trying to hedge against malware. The use case for me is when I'm running agents directly based off of prompts that I give them and asking them to modify some files. If I wanted a solution for running code I didn't trust, I wouldn't rely on what I wrote, because that's not the intended use case at all.
I went with Docker because history has taught me that new isolation strategies _will_ have escape bugs at some point, and I’m distrustful that the LLM can’t find one if it wants to.
> Probably the biggest (and least hardened) feature that I spent time on in mine was trying to figure out how to allow arbitrary GUI apps so that I could run agents in it via Zed.
I actually have code for this if you want to use/fork/borrow it. TLDR, mine pretends to be an ACP agent so you run Zed on the host, but under the hood that binary is just creating a Docker container with your image and agent, copying files, etc, and then proxying ACP messages via web socket back and forth. Except for the built-in ACP read/write file and shell endpoints. Those get executed inside the container by the proxy by default, though there’s a config option to pass either or both through to the host.
It does wrap them in protobuf and there’s some router-like stuff so you can add your own non-ACP messages between the two ends of the proxy.
main also has experimental wasm plugin support in the proxies so you can block prompts/tool calls/whatever in an agent-independent way, or add RAG that works for every agent in the world, or whatever.
I've never worked in adtech (and hopefully never will), but the first idea that would pop into my head for trying to track people who sequestered their cookies like this would be to have the sites who put in the third-party cookies include the IP address. If only one Facebook account is known to log into that IP, you know who it belongs to, even when there's no active login.
I imagine that people actively trying to make money off stuff like this would have come up with plenty more ideas than just this one.
I made a browser extension that opens every tab to a proxy with a new ip address. I have a /25 at home that will keep my browsing more or less anonymous these days. I have another 200 ip addresses available through my colo, but a non residential address limits a lot of sites
Yeah, this is just standard third-party cookie functionality, which has always been sketchy. It honestly seems like it was only possible by accident; browsers have long prevented sites from reading cookies from other domains, but it seems like the people working on early specs might not have considered the ramifications of being able to set cookies for domains other than your own. A couple decades ago it might have seemed like no one would have any reason to set a cookie they couldn't read.
Maybe 3 decades ago people had excuses, but the latest decade of cookie abuses have been designed by people who not only knew better but who took that better world into account as they buried it away from the general public. The fact that half a million developers think CORS is a server security measure isn't an accident.
I worked for AWS for several years, eventually leaving during the big "return" to office push because my geographically distributed team working on a product that didn't exist back before the remote period was told we had a month to either move to one of three cities where our team was allowed to work out of our transfer to a team in our area. My wife had an autoimmune condition that would put her at risk if I commuted, so I asked my manager what processes there were for exemptions, but he literally hadn't been told anything by the higher-ups and that as far as he was aware, there were no processes defined at all and we'd have to just try to talk with HR and upper management to try to figure something out. I didn't think that the company expecting me to rush to figure it out when they were the ones who put an artificially constrained timeline on us to either abandon what we had been working on it uproot our lives, so I ended to just giving my notice a couple of days later instead.
My point here is that companies that love to have extremely formal processes around for to handle technical things are not necessarily less likely to have extremely arbitrary decisions without any process for recourse defined when it comes to how employees get treated. If someone describes a technical process that they say they use for everything that a literal reading seems pretty dubious in regards to things like burnout or micromanagement, I don't think it's that crazy to recognize that the reality is probably at least as bad as the obvious implication. Sure, they're not directly saying "I overwork my employees by imposing short deadlines on everything and I nitpick their processes if they different from my own", but there are enough managers who do act this way that it's kind of hard to think someone who cared about being perceived as saying that wouldn't go out of their way to clarify where the nuance is if it truly exists.
reply