I’m cautiously optimistic that LLMs have a role in addressing that asymmetry to the side of good faith actors.
Gish galloping bad faith trolls aren’t new. LLMs shape their BS into fluffy BS that isn’t particularly more effective. But now, We Have The Technology, refuting a pile of poo semi-accurately should be cheap (or at least getting cheaper).
I don’t need an LLM on my phone that can do tax law in Georgia the country. But an “AI Assistant” that could highlight logical fallacies, shifting goalposts, non-responsive dialog, rhetorical obfuscations, etc, would be useful online, at the bar, and work (ie when HR tries to “HR” you, but also is lying and obfuscating about it).
We already have models and people that bullshit. Maybe refutation models are the cure… Chinese needle snakes to catch the lizards, Gorillas to catch the snakes…
This is a horrendous take. The only thing this is going to do / is already doing is increasing people's creation of their own reality bubble. LLMs are not some source of objective truth, they will inevitability lean towards reinforcing either (1) prompter's position, (2) the model trainer's position, or (3) the statistically average position, none of which are guaranteed to be logically correct. But people do take them as objective truth, so now we have a bunch of fucking morons going around saying "see, ChatGPT says so, I'm right!".
I just checked my Github settings, and found that sharing my data was "enabled".
This setting does not represent my wishes and I definitely would not have set it that way on purpose. It was either defaulted that way, or when the option was presented to me I configured it the opposite of how I intended.
Fortunately, none of the work I do these days with Copilot enabled is sensitive (if it was I would have been much more paranoid).
I'm in the USA and pay for Copilot as an individual.
Shit like this is why I pay for duck.ai where the main selling point is that the product is private by default.
It would be interesting to know whether that rule was onerous enough in practice that they had little choice but to break it in order to do their jobs effectively. They were responding to an emergency, seconds count, and they believed they had clearance from the controller.
> Deciding to change policies to effect the recommendation isn't their role.
And if it was the role of investigators to change policy, then there would be enormous pressure from industry to reach convenient conclusions, poisoning the investigation process.
In both cases, the controller's fate was grim. Peter Nielsen (Überlingen) was murdered by a relative of a crash victim. Robin Lee Wascher (LA), whose own parents had died in an earlier air crash, was crucified in the media and never worked as a controller again.
Both precedents are applicable, because the Laguardia controller is also going to be savaged.
Yes! But every news organization is leading with "I messed up." And the US President commented "They messed up", though it's unclear who that was in reference to.
Humans have a powerful need to affix blame and punish individuals. On the internet, you are forever the worst moment of your life.
We set air traffic controllers up to fail, and then when something goes wrong we torture them until they die, and then torture their memory after they die.
By using the role name rather than proper name, I'd hoped to spare HN from a tangent like this. My point doesn't rest on the nature of single individual, but instead applies to a human tendency. Politicians and press both play to the base impulses of a mass audience, unlike the NTSB. This is not the first time that a politician has scapegoated individuals when systemic failure occurs.
I actually can't remember or imagine another POTUS even getting to a level of specificity required to scapegoat an individual for something like this. The usual (and correct) answer is to say: "We don't know yet what happened, but there will be a full investigation and we will make the changes necessary to prevent it from happening again."
Pretty easy!
It doesn't serve us well to act like this administration is anything other than extremely aberrational.
Look, if you were to review my comment history you would have no doubt about where I stand on the current administration.
But scapegoating any single politician for the systemic problems of aviation is as unhelpful as scapegoating the controller for the crash at Laguardia.
I’ll scapegoat a single politician. Ronald Reagan - he owns 100% of the responsibility for the current state of things when he refused to negotiate better working conditions in 1981. The entire US is still feeling the aftermath.
This is not true. Aviation in the US has problems because of the tendency for safety regulators to do CYA when making decisions instead of adopting new technology.
Leaded gasoline? Illegal to use in the US - unless you're putting it into an old plane, where it's not likely legal to put unleaded in.
ATC? Done with old radar screens and physical cards.
Ground Control? Someone has to be standing in the tower with a pair of binoculars.
The US has an extremely safe aviation system, but the price for that safety has been technological stagnation. If I spend $70k on a small airplane, the best that'll get me is a 1975 Piper with a lawnmower engine and analog gauges. Replacing those with digital instruments will run ~$20k - the instruments themselves are only $7k, but the regulatory burden is quite pricey.
Reagan didn't do the US any favors when he treated ATC as disposable, but the truth is that the volume of flights has increased enormously and the job of ATC has gotten much harder while at the same time controller staffing has been screwed by budget fights in Congress and a couple years of one very misguided DEI policy.
The US needs to automate more of ATC. Human beings should be dedicated to emergencies, not issuing the exact same clearance 300 times a day.
It is absolutely true. I stated that Reagan is the reason that ATC are overworked and underpaid.
You proceeded to list a bunch of things that have absolutely nothing to do with ATC being overworked and underpaid.
"Automating more of ATC" would change absolutely nothing about the fact they're overworked and underpaid, there would just be fewer controllers with the same workload because they lost all ability to collectively bargain with Reagan.
Name an industry that has automated, and the end result was they kept the same number of employees, but paid them more and reduced their hours. Oh, and it can't be a unionized industry. I'll wait patiently wait for that list.
> Name an industry that has automated, and the end result was they kept the same number of employees, but paid them more and reduced their hours. Oh, and it can't be a unionized industry. I'll wait patiently wait for that list.
I'm not providing that list because it's stupid. ATC is not a jobs program; it's a profession that exists to solve a problem. The goal is not to pay ATC more, the goal is to safely manage air traffic at a reasonable price.
There is a ton of low hanging fruit because ATC is done today via phone calls and analog radio despite digital radar and mandatory transponders. It would substantially reduce controller workload, because important yet brainless tasks like "don't issue a clearance to cross a runway with landing traffic" are trivial for a computer but require the same amount of synchronous focus for a human as managing an emergency landing.
Clearances to cross a runway are given by someone with a radio and a pair of binoculars right now, which is how this was possible. With another few controllers it would have been less likely.
With a few traffic lights and computers controlling them? This wouldn't be possible at all, because the controller could focus on the emergency and the rest of the traffic could just run as normal.
The number of flights in the US is enormous and still growing. ATC, as a job, really sucks because you have to spend years in school and then commit to a career where the government can just decide where you're going to live on a whim (no, a union would not fix this, because everywhere needs ATC but not everyone wants to live everywhere). You have criminal liability if you make a mistake and while you can make six figures, it's very hard to make as much as you would at a similarly stressful and intellectual job because anything in the private sector that's this critical just gets automated ASAP.
I have a pilot's license. I can tell you with certainty that even when ATC is staffed for conditions they still make mistakes fairly often. That's just the nature of the problem no matter how much you pay them or how many controllers you hire. When you're landing a 200mph jetliner every 60 seconds there is too much room for error in a human brain.
So public sector unions can do no wrong? Can never ask for too much? The public, and by extension, the politicians that they elect, is never allowed to question or refuse their demands?
Your belief is that no other politician in the next half century has had any responsibility for the state of ATC today? No politicians in that time could have increased their pay or increased recruiting and staffing numbers?
I didn’t see anyone scapegoating him for anything other than engaging in direct personal attribution which is counter to aviation safety culture, basic leadership principles, and minimum decorum standards ¯\_(ツ)_/¯
Agreed. Respect and decorum are gone with the most recent POTUS. It's not okay to ascribe this aviation incident to the ATC controller. However, it is fully okay to call the POTUS and staff out for attacking so many individuals, at such a deeply personal level, over issues that are clearly systematic and that have clearly gotten worse under current leadership.
Sure but most of his predecessors knew enough to not weigh in beyond regret for the tragedy and loss of life until after the investigators did their job.
The groups I ran were scheduled during lunch. Technical management would look the other way if we ran over time or if people spent a certain amount of their work day reading the material.
Even if you have enlightened technical management, it's helpful if you don't force them to spend political capital justifying groups like this. Getting our enlightened CTO to spend a few hundred dollars on books was easy when we were a startup. Once we got acquired, making that argument to unreceptive higher-ups wasn't worth it for anybody.
I gave a long post at the top level about running a book-focused reading group at a company, but your group sounds more like a Papers We Love[1] chapter.
I used to co-host the San Diego chapter of Papers We Love[2], and here's my secret sauce: I offered to meet with every presenter in advance for a dry run of their presentation. Probably two thirds of the presenters took me up on the offer.
For the group and the presenter, going through a dry run had the positive impact you would expect on presentation quality.
The benefit for me was that I got one-on-one discussion/learning with a wide variety of people passionate about a broad range of papers, and I also got to go through the material twice. So I learned much more and retained it better.
The dry run idea is really smart. We've done something similar, where we had Niv Dayan[1] lead a session on Diva[2] (before it won Best Paper at VLDB 2025!). I had worked with him in the past and thought it would be cool to have him present to the group. Having the author in the room completely changed the quality of the discussion. Most of our sessions right now aren't presenter-led, but I'd like to do more of that.
I ran reading groups over several years a medium-sized startup I worked for and for an open source community I was a part of. The groups were targeted not only at engineering staff but also semi-technical positions such as product management and low-skilled data pipeline specialists, with the aim of building bridges between departments and internal recruitment.
Running those reading groups was basically how I acquired a lot of knowledge I didn't get because I didn't major in Comp Sci as an undergrad.
Books/MOOCs we went through:
* Pro Git (Chacon) [three times]
* Programming Language Pragmatics (Scott)
* Calculus first semester (Coursera/OSU/Fowler)
* Eloquent JavaScript (Haverbeke)
* Learning Perl (Schwartz/Foy) [two or three times]
* Coding the Matrix (Coursera/Brown/Klein) [didn't finish]
* Object-Oriented Programming: An Evolutionary Approach (Cox)
* Learning Core Audio (Adamson/Avila)
The level of commitment from the participants was mixed. Nobody came for the free lunch (although arranging free lunch was essential in getting people to show up). Many people had ambitions that outstripped their commitment. But there were plenty of people who took advantage and learned tons.
One fellow, who I feel immensely fortunate to have known, went from zero experience to arguably our most productive IC within 2 years, and eventually landed at a FAANG. He also took a turn leading the discussion group.
Another participant was a Product Manager who became much better able to communicate with Engineering after improving her understanding of programming fundamentals.
The sessions were generally organized around questions that people would bring about the material — my motto was "we'll struggle through together". I preferred questions posed by others but always had enough of my own to fill space.
The tips I would have for people running such groups:
First you need a discussion leader. You need someone who can get the group unstuck and who knows the material. It's the same dynamic as having a good TA for a college discussion section.
Second, remove all friction. Corporate won't buy books? Screw it, I bought 'em myself. Corporate won't arrange for lunch? Screw it, I bought it for everybody. (Our CTO was highly supportive but once the company got acquired we couldn't get budget approved anymore). The total outlay I made leading the group was a few thousand dollars — far less than I would have paid for formal courses where I would have learned less.
Wow, Cox's original Objective-C book? Interesting historically, but it's hard to imagine it was of much pragmatic use, even if you're working in Obj-C as your day-to-day. Still, it's an interesting artifact of the early OO age, and the metaphor of libraries of objects as integrated circuits was interesting.
Yes! And I wouldn't have picked it, but I went along with it when someone else wanted to read it.
The "software IC" metaphor may not have caught on, but these artifacts of early experimentation, competitors against what evolved into mainstream OOP, are many times more interesting to read than the Nth "OOP sux" article.
EDIT: I just went back over some of my notes... Cox predicted that companies would compete to provide different implementations of common interfaces ("software ICs"). In restrospect, that didn't happen but Cox's prediction was wrong in such an interesting way: software orgs rather than customers assumed control over interfaces.
> The amount of energy needed to refute bullshit is an order of magnitude bigger than that needed to produce it.
https://en.wikipedia.org/wiki/Brandolini%27s_law
reply