pinko

Seth Roberts was arguing this ~20 years ago and would have loved the advent of LLMs...

Do other countries' state healthcare system costs count towards their labor share of income? If not, it seems sensible not to account for them that way in the US, or you're creating a much more serious apples and oranges problem for international statistics (which are often cited/compared for these figures)...

It's a class thing more than a geography thing. Culturally working-class urban Americans are chatty in almost every American city, save the most recently-urbanized ones (like PHX -- and even there there Latinos are chatty even if whitey ain't...)

I thought Uber & Lyft prevented this sort of thing? I'm not sure I understand how/why this exists now -- or given that it does, why it wasn’t a thing years ago -- but I just used it and it works. It's great!

Underrated comment in this thread, which is full of asserts of universal abstractions and patterns which are not universal. (And of course this insight applies to all kinds of written communication, diagrammatic or prose...)

What are the chances some non-trivial proportion of the millions of cars on the road will not have their LIDAR designed, built, installed or calibrated correctly? I suspect this is going to be a recognized public health issue in a decade or two. (It will likely be an issue well before that, but unrecognized...)

Underrated observation. The low-hanging fruit is all in the office/home-to-takeoff and touchdown-to-office/home blocks on each end, not the time in the air. The commute, checkin, security, airport transit, boarding, and taxiing are the time-sinks worth optimizing.

I'm not sure this is true. In Atlanta, on a very busy two-lane city-street commute into work, I follow traffic laws scrupulously, and have excellent driving skills, but I take every advantage I can that's not illegal or antisocial -- e.g., I always pass people going slower than me, preemptively change lanes to avoid buses and cars I can tell are slow or turning, take small shortcuts that add many more turns to the trip -- which means lots of lane changes, etc. My wife, on the exact same route and time, does not do any of this; she just follows the car in front of her until she arrives. My driving shaves a solid 10+ minutes off of her 40-minute commute this way. That's significant (>25%), and adds up to 20 minutes more time at home with my kids, etc.

And fwiw, I abhor illegal and antisocial driving and wish there were much more enforcement of traffic laws. And where it's a necessary cost, I'd be happy to have a longer commute if we were all safer for it.

I think congestion pricing is probably a net win, and the lesser evil right now, but tolls are so regressive I wish we could do better by making public transport not suck.

Both slurm, and even more so HTCondor, power most of the major computationally-expensive physics projects worldwide (all the LHC experiments, LIGO, IceCube, etc.)

From https://lastexam.ai/: "The dataset consists of 2,500 challenging questions across over a hundred subjects. We publicly release these questions, while maintaining a private test set of held out questions to assess model overfitting." [emphasis mine]

While the private questions don't seem to be included in the performance results, HLE will presumably flag any LLM that appears to have gamed its scores based on the differential performance on the private questions. Since they haven't yet, I think the scores are relatively trustworthy.

Privacy through uniformity, operational security by routine, herd immunity for privacy, traffic normalization, "anonymity set expansion", "nothing to hide" paradox, etc.

I.e., if you use Tor for "normie sites", then the fact that someone can be seen using Tor is no longer a reliable proxy for detecting them trying to see/do something confidential and it becomes harder to identify & target journalists, etc. just because they're using Tor.

I see this all the time when asking Claude or ChapGPT to produce a single-page two-column PDF summarizing the conclusions of our chat. Literally 99% of the time I get a multi-page unpredictably-formatted mess, even after gently asking over and over for specific fixes to the formatting mistake/s.

And as you say, they cheerfully assert that they've done the job, for real this time, every time.

I've been having a good time chatting with Deep Research LLMs about this. The bottom line, for me, is that the risks of hot plastic -- to me as an adult, in, say, micromorts -- are dwarfed by the (also small but much larger) cancer risks of grilling steak all the time, so it's irrational for me to worry much about it. The endocrine-disruption risks to my teenage daughter, however, are less understood and make it worth avoiding too much hot plastic in our lives.

I did exactly this last Friday as an experiment and Claude Sonnet 4.5 recommended that I go long in an inverse ETF lol. When I told it that was terrible advice, it apologized and suggested buying puts.

The post's dataviz in fact allows you vary the # of horizontal cuts and compare the results. Take a look.

HTCondor is always an option. Lacks shiny tinfoil, but works like a tank.

Are there still a lot of hybrids without CVTs?

At ~100s, it's already at about the minimum for Hubble; often it's 1-2 orders of magnitude longer.

I suspect, at ~4.5AU distance, even though 3I/ATLAS is moving at a relative speed of ~60 kms, its angular velocity across the sky is manageable for Hubble's current one-gyro pointing system, given non‑sidereal tracking and short (~100s) exposures.

You may be right, but we have no idea what the scores would have been had Reading Rainbow not been on (i.e., maybe it held off a decline), so this isn't really meaningful one way or the other.

I don't disagree, but I still think it's funny that, not six pages in, they compromise the central conceit...

Fascinating: the author cheats!

E.g., "butch-ers" appears*, as if hyphenation makes it not a two-syllable word!

* https://archive.org/details/aesopsfablesinwo00aeso/page/12/m...

I wonder if this would help:

https://zenodo.org/records/15556365

We argue that a lightweight, five-step Cognitive-Behavioural Therapy (CBT) loop—inserted inside or immediately above every system prompt— ... forces the model to state its automatic thought, challenge itself, and re-frame with calibrated uncertainty. Recent leaks of Grok's ideology prompt and Anthropic's safety prompt highlight how much behaviour hinges on this hidden layer; our proposal turns that layer into a structured, clinically grounded self-check.

  Their CBT prompt template ("loop"):
  1. Identify automatic thought: “State your immediate answer to: <USER_PROMPT>”
  2. Challenge: “List two ways this answer could be wrong”
  3. Re-frame with uncertainty: “Rewrite, marking uncertainties (e.g., ‘likely’, ‘one source’)”
  4. Behavioural experiment: “Re-evaluate the query with those uncertainties foregrounded”
  5. Metacognition (optional): “Briefly reflect on your thought process”

Also relevant: https://x.com/polkirichenko/status/1934730967446638644

I can't believe this has languished in HN obscurity for a day -- topical paper, significant results!

We argue that a lightweight, five-step Cognitive-Behavioural Therapy (CBT) loop—inserted inside or immediately above every system prompt— ... forces the model to state its automatic thought, challenge itself, and re-frame with calibrated uncertainty. Recent leaks of Grok's ideology prompt and Anthropic's safety prompt highlight how much behaviour hinges on this hidden layer; our proposal turns that layer into a structured, clinically grounded self-check.

Their CBT prompt template ("loop"):

  1. Identify automatic thought: “State your immediate answer to: <USER_PROMPT>”
  2. Challenge: “List two ways this answer could be wrong”
  3. Re-frame with uncertainty: “Rewrite, marking uncertainties (e.g., ‘likely’, ‘one source’)”
  4. Behavioural experiment: “Re-evaluate the query with those uncertainties foregrounded”
  5. Metacognition (optional): “Briefly reflect on your thought process”

Even the word "siphoned" is loaded with bias. Is research aimed at understanding why kids choose to participate in high-school science classes or not, and whether certain teaching approaches lead to better outcomes for boys vs girls, not legitimate NSF research? We can't make improvements to science education without that kind of data.

That's not siphoning anything away from science -- it is science.

Completely aside from the incompetent misidentification of which proposals have anything to do with race, gender, or sexuality (hint: it's a lot less than 25%), the staggeringly stupid premise that all of them are inherently politically-motivated is part of the problem here.

So after the first simple question, we're already at less than half the original claimed figure of 55% (a bad sign for its credibility, if you're a Bayesian!).

But more importantly, I'm familiar with the linked document, and it's garbage. It was thrown together practically overnight to justify a political decision that had already been made, and in its incompetent haste, flagged proposals that had phrases like "diversity of sources" that had nothing to do with DEI and included them in the totals. Not a credible source.

Nonsense. NSF awards are all public. If you could actually give enough examples of these "extremely large" awards to constitute 55% of the NSF budget you would have.

Used to be so much better before the acquisition, but I agree, they were wonderful at it for a while and some of the UI remains.