I wonder if this will actually be why the models move to "neuralese" or whatever non-language latent representation people work out. Interpretability disappears but efficiency potentially goes way up. Even without a performance increase that would be pretty huge.
I just can't find myself summoning the energy to be mad about markdown. It's good enough for like 99% of the things I use it for. Sometimes I get annoyed at specific extension support or whatever when I realize I shouldn't be using markdown for that task.
> The devices are either dangerous, or they're not
That's not actually how it works though, it's all a risk and percentages. Nobody says "driving is either safe or it's not" or "delivering a baby is either safe or it's not"
Correct, but I agree with the parent that this is a dubious case to apply that reasoning.
To make it clearer, imagine another context: "It's dangerous for a passenger to have a gun on board. Therefore, we're strictly limiting passengers to only two guns."
Like, no. The relevant sad case is present with one gun just as with two.
Of course, what complicates it is that these power banks present a small but relevant risk of burning and killing everyone on board. So yeah, you might be below the risk threshold if everyone brought two, but not three. So it's not inherently a stupid idea, but requires a really precise risk calculation to justify that figure.
That's not really how risk is managed in aviation. ICAO will have made a list of all possible ways a power bank could create a hazard. Then for each failure mode, they'll come up with two numbers: probability, and severity. There's a formula to combine those two numbers into a single risk score. Any risks over the acceptable threshold (varies depending on the circumstances and I can't remember what it is for human-rated transport) must be mitigated.
A mitigation is anything that reduces the probability or the severity of a risk. There are different categories of mitigation, some of which are more robust than others. Once the risk score moves below the acceptable threshold, the risk is satisfactorily mitigated.
Example: Rapid depressurization. Without mitigation, the risk of rapid depressurization is unacceptably high. So we mitigate the probability by requiring sensitive inspections for metal fatigue, and we mitigate the severity by providing oxygen masks, a standard flight crew procedure for making an emergency descent, and regular training on that procedure. (Plus a bunch of other things I'm not thinking of off the top of my head.)
Assuming ICAO did their due diligence - and I don't have any reason to think they didn't - they would've assessed the probability and severity of all of the ways a consumer power bank might fail. That analysis is the rationale for both the number of power banks allowed on a flight and what you're allowed to do with them. And yes, they will have considered the probability of people not following the rules (which is the reason, btw, that airplane lavatories have enormous "no smoking" signs right above an ash tray).
This is the first decent answer, which I appreciate. And while my comparison to a bomb may have been over the top, I don't think a comparison to shampoo is fair either. And in any case, I'm not so sure whether the limit on toiletries is all that sensical either.
I’m curious if you think viewpoints have also gotten more extreme in this period. It feels like the gap in political ideologies has widened a lot since I was younger.
The gaps in views are certainly widening in societies in general, but my point wasn't exactly that. It was more about differences which always existed. When I was young I was often dumped into large family gatherings which lasted days (birthdays of (grand)grandparents, funerals, weddings etc). I had to practice handling cousins etc who might had very different family backgrounds than me since very early age. We had to find things we had in common and accept our differences. We learned that differences are manageable. It's not common nowadays. Many people don't have relationships with relatives at all and kids don't meet another kids with different background until school. And even then distance is kept often because of overprotective parenting.
I agree with where you're coming from. I think its the continuing rise of individualism to a degree; sports team mentality politics/the internet/phones/social media have likely accelerated it but feels to me we'd been moving that way even prior to that. There's probably something there, too, about how more wealth gives people an out to not have to engage with broader society in a way where lower wealth or less unequal distribution of wealth doesn't.
Margaret Thatcher got quite a lot of stick for her quote about how there is no society and its just individual people that look out for themselves first and foremost before anything. But I'd say there's more truth to that than I'd want to believe. What stops that from being true to my mind is people being very intentional about creating a society by engaging with others.
I'm always interested in people's stories about adult friendships or loneliness crisis etc. and what I tend to see is that when people start being intentional about engaging, it usually ends up with them finding new friends. It's just easy to sleepwalk, with everything going on, into not engaging. Newish parents are very apt example of this
In my twenties my friend group had a gay jewish law student and an actual nazi, and everything in between. We all had the same sense of humour and didnt take ourselves seriously. The two guys specified would have deep and long conversations (not debates, not arguments) intonthe early hours while the rest of us fell asleep or left due to next day obligations. The only controversy I recall was when I said Conan the Barbarian is better than Lord of the Rings.
Yeah as long as the chatbot is empowered to fix a bunch of basic problems I'm okay with them as the first line of support. The way support is setup nowadays humans are basically forced to be robots anyway, given a set of canned responses for each scenario and almost no latitude of their own. At least the robot responds instantly.
Nice, I like the idea. It sounds like qualitatively you haven't had any performance regressions while doing this, but have you tested it at all on any sort of benchmark or similar eval? I'm curious how well the actual system performs with less context like this. I mean it's possible it actually improves...
Another interesting thing here is that the gap between "burned out but just producing subpar work" and "so crispy I literally cannot work" is even wider with AI. The bar for just firing off prompts is low, but the mental effort required to know the right prompts to ask and then validate is much higher so you just skip that part. You can work for months doing terrible work and then eventually the entire codebase collapses.
First, I agree with most commentators that they should just offer 3 modes of visibility: "default", "high", "verbose" or whatever
But I'm with you that this mode of working where you watch the agent work in real-time seems like it will be outdated soon. Even if we're not quite there, we've all seen how quickly these models improve. Last year I was saying Cursor was better because it allowed me to better understand every single change. I'm not really saying that anymore.
reply