abeppu

In the early 2000s I saw a gubernatorial candidate talking to an audience of over achieving high school students and when asked about what he would do to make college more affordable. He told a story about how he waited tables to pay for college.

Totally not on his radar: changes in tuition over time, portion of budgets of public colleges that were coming from state funds (his tuition wasn't free but was close), vs changes in median hourly wages over the same period.

Very clearly on his radar: those students were not gonna donate to anyone's campaign regardless of what he said.

Before the IPO there was much concern that the changes in rules to allow fast inclusion in indices like Nasdaq-100 and Russell 1000 would mean that there would be an artificially elevated price. Was that effect just over-estimated? Or should SPCX have dropped even further without that support?

Or if it had any kind of measurement across a range of tasks

I think this is silly. This is a product designed to deliver nicotine, and that's going to continue to be the important part for health considerations. Had the FDA become concerned about the pouch material, and a different pouch material had been swapped in, it would still be a product designed to deliver nicotine.

If it's used in a way that gets people away from smoking, maybe it can have an impact of decreased cancer risk. If used by people who weren't going to smoke, then maybe the impacts to blood pressure, heart rate, and heart disease risk are going to dominate.

The "maybe you accumulate more microplastics" impact is a rounding error.

... would there really not be regulations on giant unmanned ships? That seems concerning, though I could certainly understand international / maritime law having a gap.

Aside from the funding mechanism being a premium that everyone pays, the other part of the motivating story here that seems downright silly is:

- the patient was _forced_ to take an ambulance to another hospital, but not to be treated for anything in particular.

- the post goes into how the EMS system is expensive b/c you need trained paramedics and expensive medical equipment and such -- but sometimes they really do just need to move a patient, not treat anyone in the field or en route. Saying that you pay for the paramedics even when you're effectively just being moved between facilities _also_ seems wasteful.

Yes, we should share the cost. But once we all share the cost, maybe we should try to spend the public dollars effectively by only using ambulances when they're needed, and distinguishing between "transport patient who can't sit upright and buckle a seatbelt" and "try to stop their arterial bleeding as you speed towards the hospital".

I mean, I don't want to eat it either. But the point of generalizing the optimization framework is that it applies to other situations with the same shape.

Maybe related questions:

- if you have the resources to run just one online learning model, but can't stash/restore state to just continue a prior model where you left off, how do you find the best online-learned model?

- you have space in your garden for only one large fruit tree. You have an heirloom apple tree currently and it produces decent but not amazing fruit. How do you choose whether to replace with something else, given that a replacement may take years to develop the root system needed to be really productive?

- (more loosely) you're in career track X and have advanced over years, the work is ok but could be better. How should you decide whether to retrain as something different?

I think there is an optimization question buried here. In tech lots of people have experience with A/B tests, which function on the assumption that you have a stream of fresh sessions which are independent. Multi-armed bandits, Thompson sampling etc give us frameworks for generalizing this towards finding the best option among a finite set of candidates, if goodness is fixed over time. This is kinda the opposite end of a spectrum: you get to run one policy at a time and the whole premise is that goodness is heavily state dependent. How do you decide whether to keep going with your current policy vs when to start over with another?

Sure, the soup is good ... but is it the best they could have after 52 years? By committing to maintaining one pot for so long, they pay the opportunity cost of not being able to explore related long-lived methods. If there's a different recipe that surpasses this one after only one year of simmering, they'll never find it.

At first I thought this might be related to the secretary problem, but of course if after 50 years of recipe B, you have the option of switching back to recipe A if it's better.

I did grep through the paper, and ... no not really.

"Urban" and "rural" never appear. Among uses of "control", for their own work they have controlled for sex and race. Their Table 2 and Figure 2 are the same data displayed different ways and from the same analysis. In their mentions of other work they note related work on Europe that controlled for religiosity, education and income, and socioeconomic and attitudinal factors.

Their table 1 actually shows that the p-value for religious attendance, education, two race variables , and birth year, were all tied with the politics variable, and showed that education, and both race variables had larger-magnitude estimates and z-values.

After reading some of the paper, while I still think there's a relationship between political beliefs and number of kids, I don't see that their methods are convincing that this is more important than other such correlations.

I didn't read the full paper only the abstract, but bc the political spectrum is correlated with education level or urban vs rural or multiple other factors that deeply impact what it is like to have kids. Of urban college educated professionals, do the conservatives still have more kids than the progressives? Or does the opportunity cost on careers plus the high cost of childcare plus the cost of living space impact conservatives and progressives alike? And among rural people without a college degree, do progressives still have fewer kids than their conservative neighbors?

When a bunch of demographic factors are all not just correlated but are linked to challenges in raising kids, it seems like elevating a single one of those factors is selective framing

I think the author basically decides that "productivity" is not defined but "effective" is. The problem with productivity is that most of us aren't directly producing a thing that is of independent and immediately clear worth. I would claim the same is true if "effectiveness" -- what effect is being achieved, and is it actually worthwhile on its own or is its worth contingent on a bunch of other work and conditions some of which you don't control or observe?

NLRB documents from proceedings between Atlassian and Denise Unterwurzacher, which yesterday produced an Administrative Law Judges Decision. This may be relevant to many tech workers both for (a) detailed and perhaps very familiar examples of what is "Protected Concerted Activity" and (b) findings about what are unlawfully overboard confidentiality and Nondisparagement clauses in a severance agreement.

Ok here's maybe a dumb or maybe a crazy question or maybe not:

- the Fed has price stability as part of its dual mandate, and in its normal operation does this at the level of manipulating the money supply for everyone, through changing the interest rates given to large banks (IIUC, I'm def not an expert here)

- the FTC has as its primary mission anti-trust enforcement and consumer protection. During the last administration, the FTC tried to be more aggressive largely through legal action with specific firms.

Price stability and anti-trust enforcement are related. Would both goals be better served if a single public body with high independence could use both tools and levels of targeting? E.g. rather than being hit with a small-ish fine, should firms that collude to manipulate prices, or firms that consolidate to the point of approaching monopoly, be penalized with higher interest rates on all their financing?

And notably, we used to have a somewhat progressive corporate income tax which, at least on paper, provided a quantitative disincentive against too much consolidation. Sometimes the merger of A and B would pay a higher rate than A and B separately. And we gave that mechanism up.

There is something a bit ironic that the evidence to refute the "more research is better" position is selecting (cherry-picking?) a particularly bad study. Are there better ways to measure whether the population of studies using these databases has declined as a whole, rather than just saying there are more of them and at least some are pretty bad?

Or people that choose organic are also choosing other things that they think are healthy, and they're right at least some of the time.

I think #2 risk being incoherent unless you define things very carefully.

"Illusion" ordinarily means there's someone with a subjective experience which creates incorrect beliefs about the world. E.g. I drive on a highway in summer, I see reflections on the road, I momentarily believe there is standing water, but it's an illusion. What does it mean for the basis of subjective experience to be illusory? Who experiences the illusion?

Pain isn't a real thing any more than an IEEE float is a real thing. A circuit flips bits and an LED shows a number. A set of neurons fire in a pattern and the word "Ow!" comes out of someone's mouth.

But we don't think the circuit has an experience of being on or off. And we _do_ think there's a difference between nerve impulses we're unaware of (e.g. your enteric nervous system most of the time) and ones we are aware of (saying "ow"). Declaring it to be "not any more real" than the led case doesn't explain the difference between nervous system behavior which does or doesn't rise to the level of conscious awareness.

RF was nowhere on my radar.

cue rimshot

Waymos pull over into bike lanes all the time for pickups and drop-offs and that’s neither legal nor safe.

While perhaps drop-offs are often relatively quick (though perhaps more risky; see the dooring accident description in the article), I'm also really annoyed by Waymos waiting and blocking for pick-ups, which can be multiple minutes.

This is also the original way variational methods pick a parameterization of a model of known architecture which best matches some distribution which generated data but is not otherwise compactly expressible.

The 0.1% thing ... Is that even the right label? I'm guessing one in a thousand people globally isn't using these mechanisms. The article spends some paragraphs on the world's richest person and his company's tax strategy. Is the millionaire next door quietly doing these things or is this about billionaires in which case it's more like one in a million.

ok so it seems pretty bad that they changed the index rules both to allow spacex in early and the wonky weighting stuff. But if one already has index-based things that are likely to be captive on the wrong side of this, and one wanted to benefit or at least balance out, to confirm my limited understanding the goal would be:

- buy shortly after the IPO, ideally less than 15 days

- and sell less than 6 months later when lockups would end and insiders are set to cash out?

I think the "Leave them Behind" section at the end sort of ignores the whole "they will ruthlessly copy your material, and put aggressive extra load on your server while repeatedly stealing your work" dimension.

You can try to avoid consuming AI-generated material, but of course part-way through a lot of things you may wonder whether it is partly AI-generated, and we don't yet have a credible "human-authored" stamp. But you can't really keep them from using your work to make cheap copies of you, or at least reducing your audience by including information or insights from your work in the chat sessions of people who otherwise might have read your work.

Microsoft bought it for OpenAI only, to train Copilot on the vast amount of code.

I think this gets the timeline wrong. Microsoft acquired GH in 2018 and started the partnership with OpenAI in summer 2019.

I'm sure there was some strategy to extract value from it that wouldn't serve its users but I think OpenAI was not initially meant to be the beneficiary.

While I agree that we keep reinventing stuff, in CS doesn't the ease of creating isomorphisms between different ways of doing things mean that canonicalization will always be a matter of some community choosing their favorite form, perhaps based on aesthetic or cultural reasons, rather than anything "universal and eternal"?

I worked for a startup that used clojure and found it so frustrating because, following the idiomatic style, pathways passed maps around, added keys to maps, etc. For any definition which received some such map, you had to read the whole pathway to understand what was expected to be in it (and therefore how you could call it, modify it, etc).

I think the thing is that yes, `[a] -> [a]` tells you relatively little about the particular relationship between lists that the function achieves, but in other languages such a signature tells you _everything_ about:

- what you need to call invoke it

- what the implementation can assume about its argument

i.e. how to use or change the function is much clearer

Do publishers really have fact-checkers? My understanding was that support for authors is now relatively minimal, even for established authors, and no one really has the time or resources to second-guess everything an author has claimed. I take as a key example Naomi Wolf learning after her book was "done" that a significant chunk of it was based on a misunderstanding of an admittedly confusing 19th century British legal phrase. https://nymag.com/intelligencer/2019/05/naomi-wolfs-book-cor...

I think maybe the idea that a single author spending months or years on their research, which the publish as a single bound and polished work is misguided -- an academic trying to do similar work in multiple articles would have gotten review from peers on each article, and hopefully have not spent so much time working under a correctable misunderstanding.

This paper describes finding security related concepts and using them to steer at generation time. While this is an interesting contribution on its own, the approach could also be applied to a range of other concerns -- e.g. can we use this to steer away from performance problems? can we make llm code generation anticipate maintainability or readability issues?

ha ok that's fair maybe it needs to be sold with a package of associated tests

If people want to try untested peptides, I think society should use that as the engine to _test those peptides_. Instead of buying something that's supposed to but may not be the peptide you want, you should pay 50+k% + data and get something that has a 50% chance of being the peptide and 50% chance of being a placebo, and you're _required_ to submit a report about effects and side effects before you can get a refill.

Rather than complain about how these things have not yet gone through real experiments and are marketed as having been "studied" rather than "effective", I would love to see society use the obvious demand for some of these to actually test them.