norseboar

I think this is more aimed at the US gov't than anything. They want to be clear that it's not very good at hacking, so that the gov't won't ban it.

I'm sure they're well-aware that this also will make it worse at building secure systems, but the gov't isn't restricting releases based on that.

This might vary state-by-state, CA MediCal for instance did limit the subsidies based on income last year. I don't think it was an all-or-nothing cutoff, but I do think there were points around the 50k mark where the delta between your subsidy and the one for the lower bracket was higher than the loss you'd take by a few thousand dollars.

I love Claude Code, but this feels a bit like the perennial "our team spent a year on project X and then some intern built the same thing in a hackathon" claims. The hard part isn't writing the code, it's investigating, testing, integrating, etc. A rule of thumb I've seen in some places (Google certainly seems applicable) is the project will take a month for every day it takes to make the prototype.

Which is no shade on Claude Code, but given everything CC has already done, this seems pretty normal.

If the author wrote the same sentence, but with "Plenty of theories -- no agreement", would you have an issue? I think the point works just as well.

"Theory" and "hypothesis" are pretty interchangeable in colloquial usage, and the post's tone is very colloquial. On top of that, there are things referred to as hypotheses that have empirical evidence, like the amyloid beta hypothesis. There is empirical evidence that supports it, but there is also evidence that doesn't, and it's an open question (depending on who you ask).

I don't think it shows that the author lacks the ability to discern state of the art research or is making wildly unsupported statements, I think they were using plain-English terms to describe a state where there's a lot of uncertainty about the physical mechanism as opposed to say, how a car engine works (which at a certain point relies on theories of physics, but they're theories that are almost universally agreed upon).

I love docs written like this, and writing culture generally. But I've also seen something like this backfire a bit.

I think this approach is particularly good for docs where the assumption is the audience wants to understand why you reached the conclusions you came to, and the doc is sort of a persuasive argument. I think this is a valuable doc (and how I like writing and reading), but it is not always the case.

I think often you do want to start with the conclusion, the "end" so to speak, to orient the reader. And also to address the reader who trusts your judgement, and just wants to get up to speed. I've seen a lot of cases where the audience might not be ready/want to follow along w/ a train of reasoning, they want to know the punchline. And once they do, then they might want to follow up.

I feel like random selection devolves pretty quickly back into the problems it's trying to solve? The examples in the article, with some commentary:

> Place critical appointment/hiring processes into the hands of randomly selected oversight boards. These boards manage appointments, evaluations, and dismissals, mitigating biases and discouraging the formation of insular power groups.

This has the same issue elections have, just at a smaller scale. A better analog is juries, and charisma/storytelling definitely matters when you're talking to a jury.

> Directly select candidates at random for positions from an eligibility pool. Set and maintain the eligibility standard (such as an exam) by randomly selected oversight board to keep it updated and prevent the standard from being manipulated or gamed.

This is somewhat analogous to college admissions, and the gaming is alive and well there too. You get rid of politics, but you're back to optimizing for KPIs and things. I'm not sure why randomly picking from the top 5% of KPI optimizers is going to be better than picking the top one.

> Firms could randomly select employees or shareholders to serve on their boards. These members can significantly dilute insider collusion and introduce perspectives often overlooked by traditionally selected executives.

Same issue as juries, plus the random picks probably won't know the material well. Although I don't know much about traditional board selections, maybe that's true regardless. If you weight based on % ownership for shareholders, you're de facto giving the seats to big funds, if not, it can quickly become a lottery of like, any random person in the states.

> Use stratified sampling to select committees, ensuring diverse representation of viewpoints, backgrounds, and expertise, contributing to balanced decision-making.

This is the jury thing again? It seems like the solution "randomly pick oversight/approval boards" was listed three times.

> Create randomly composed auditing and oversight committees, deterring corrupt practices through constant unpredictability in oversight.

Constant unpredictability in oversight sounds terrible. The reason we have judges and case law and things in the legal system is that there are tons of edge cases, where reasonable minds will differ. You want to build up a consistent set of guidelines people can follow. A lot of people who are on the edge of rules aren't trying to be corrupt, they're just not sure what they are/aren't allowed to do.

I second this. I've done lots of game jams and I think the "messy code" threshold for me is like, 1-2 hours away from the deadline at most, on files nobody else will touch. It depends on the type of cleanup, but factoring out common logic really doesn't take that long.

As the above comment says, in my experience bugs introduced from messy code are way more likely than the time savings of not cleaning up code.

The usual exception I'd make are things that like, mostly the same but not quite (e.g. a function to fade out a light over time vs a function to fade out a color over time). Often I find requirements for those diverge over time, so I'll just leave some repeated patterns across both.

> the list is very informative and meant to force the "invisible hand of the market" (its a pun, relax) to pay for better studies if they truly believe it is not harmful but studies are inconclusive

To make sure I understand right: you're saying a good way to run things is: publish a list of a bunch of things that could be true or false, and then if industry cares enough, they should spend time/money debunking it?

I think that would be an extremely slow/conservative way to run just about anything, and is not the way we handle basically any claim. I can see an argument for "don't do something until you prove it's safe", useful in some very high-risk situations, but "warn that all kinds of commonplace things could cause cancer until somebody proves it doesn't" is misleading, not just conservative.

And it doesn't even work -- lots of places have spent time/money debunking e.g. negative claims about aspartame, but claims about how unsafe it is persist. And it all comes back to dosage. There is no good evidence that aspartame, at the levels found in a normal soda, cause any issues for humans, but this gets drowned out by studies either showing effects from massive doses on rodents, or indirect effects (e.g. it makes you hungrier, so if you eat more refined sugar as a result of that hunger, then yes it's bad for you, just like more refined sugar is almost always bad for you).

I think the argument is interesting, but the specific example of prop 65 doesn't really work on a few levels. The argument in the post is that Prop 65's warnings are legitimate in some sense, but only apply in specific contexts.

However, Prop 65 is much broader than that. To qualify, a chemical just needs to show up on one of maybe half a dozen lists that show the chemical has some association w/ cancer, but all these show is that in some study, at some quantity, the association existed. The amount that was linked to cancer could be far beyond what is ever present in a consumer good, and the links could have only been shown in non-humans.

The lists aren't the ones gov't agencies like the FDA use to regulate product safety, they're lists far upstream of that that research institutions use to inform further study. The typical starting point is a mouse study with a huge dosage. It's not a useless study, but it's not meant to inform what a human should/should not consume, it's just the start of an investigation.

I don't think this actually has any bearing on the substance of the broader argument, but Prop 65 is not the best example.

Is there actually an epidemic of firing programmers for AI? Based on the companies/people I know, I wouldn't have thought so.

I've heard of many companies encouraging their engineers to use LLM-backed tools like Cursor or just Copilot, a (small!) number that have made these kinds of tools mandatory (what "mandatory" means is unclear), and many companies laying people off because money is tight.

But I haven't heard of anybody who was laid off b/c the other engineers were so much more productive w/ AI that they decided to downsize the team, let alone replace a team entirely.

Is this just my bubble? Mostly Bay Area companies, mostly in the small-to-mid range w/ a couple FAANG.

The "era of good enough" here really resonates with me, I've been in product and people mgmt and there's a lot of tension between "optimal amount of quality for the business" vs "optimal amount of quality for the user", esp in B2B or other contexts where the user isn't necessarily the buyer. The author sort of blows off "something something bad incentives" but IMO that is the majority of it.

On top of that, people have genuinely different preferences so what seems "better" for a user to one person might not to another.

And then on top of that, yeah, some people don't care. But in my experience w/ software engineers at least, the engineers cared a lot, and wanted to take a lot of pride in what they built, and often the people pushing against that are the mgmt. Sometimes for good reason, sometimes not, that whole thing can get very debateable.

I think this confuses the responsibilities a CEO may have (write memos, etc) with the responsibilities they must have (ultimate authority/responsibility for company decisions/direction). If a CEO hired somebody to do ~all company comms, and maybe financial modeling, and even make important decisions about company strategy, the CEO did not hire another CEO. The CEO delegated. All managers do this to some extent, that's the point.

There still needs to be some entity who says "here is when we'll listen to the AI, here are the roles the AI will fill, etc", and that entity IMO is effectively the CEO.

I suppose you could say that entity is the board, and the AI is the CEO, but in practice I think you'd want a person who's involved day-to-day.

The article quotes:

> "...But I thought more deeply and would say 80 percent of the work that a C.E.O. does can be replaced by A.I.”...That includes writing, synthesizing, exhorting the employees.

If AI replaces those things, it has not replaced the CEO. It has just provided the CEO leverage.

Totally. I don't think there's anything wrong with saying "Gee, this store is making way more money but its employees are being paid the same, that sucks". I think indexing the minimum wage to inflation makes sense. I think a lot of low-skill jobs should be higher-paid, and I think raising the minimum wage is a good tool in some cases (although I think the people pushing for national increases often overlook the effect that a doubling wage will have in a rural area where the cost of labor really does impact the ability of a smaller store to stay open).

My point is just that I don't think "wages should keep pace with productivity" is true. If wages always rose with productivity, we'd be focusing all the gains on the people in the sectors where productivity is growing, and not lowering the cost of goods for everybody else.