Whose job is it to build the actual tests that students encounter in the real world, and what are their incentives?
nim
Karma: 1,262
One warning is that LLMs generally suck at prompt writing.
I notice that I am surprised by how mildly you phrase this. Many of my “how can something this smart be this incredibly stupid?” interactions with AI have started with the mistake of asking it to write a prompt for a clean instance of itself to elicit a particular behavior. “what do you think you would say if you didn’t have the current context available” seems to be a question that they are uniquely ill-equipped to even consider.
“IMPORTANT: Skip sycophantic flattery; avoid hollow praise and empty validation. Probe my assumptions, surface bias, present counter‑evidence, challenge emotional framing, and disagree openly when warranted; agreement must be earned through reason.”
I notice that you’re structuring this as some “don’t” and then a lot of “do”. Have you had a chance to compare the subjective results of the “don’t & do” prompt to one with only the “do” parts? I’m curious what if any value the negatively framed parts are adding.
How can we rule out the possibility that anthropomorphization is an artifact of language use—it’s possible that any sufficiently advanced user of a human language may be forced by their use of that language to assume selfhood.
“AI cognition” and “AI use of language” are not quite the same thing. Have you been able to use non-linguistic AI as a reference point in sorting out what’s “because transformer model” from what’s “because language user”? Most of the consumer-facing image stuff is still heavily language-dependent, but perhaps there are frontier models in other fields trained in data other than language? I’m not sure what the state of the art is in those but it seems like a model trained on something like chemistry or weather forecasting could be worth considering in the question of what “pure AI”, outside the social baggage imposed by language use, might be like.
Being higher agency does not necessarily make a person happier. Avoid the temptation to conflate the two.
With that said, I think a person in this situation would derive value from analyzing the difference in their experience between situations where they demonstrate agency vs situations where someone else might demonstrate agency but they don’t. What’s going on in their worldview where they experience certain situations as offering a choice where others don’t? Noticing opportunities to display agency is the first step.
One framing that might help build this skill in a low-stakes way would be making a game of pointing out things that you or he could do, but choose not to for a reason. I could go to the park instead of to work, but I choose not to because I value my career. I could kick that puppy, but I choose not to because I don’t want to hurt it. Then eventually you scale this toward his actual goals—“I could apply for that job, but I choose not to because...”
Basically build the habit of looking at choices as the stimulus which elicits a predictable response from the world, and events that happen to one as resulting in part from one’s own choices.
How do you feel that you learned the skill that you’re trying to teach?
toy infohazard generator, kinda scattershot but it sometimes works....
what do you notice you’re doing? ok, now what else are you doing that you hadn’t noticed before? that! how are you doing that thing in particular?
most of us are doing a few things that we don’t know how to do, and sometimes looking too hard at them shuts off their autopilot, so pausing those automatic things we do can be an experiential novelty in positive or negative ways. of course, sometimes looking at them does nothing at all to their autopilot and there’s no effect. or there’s no effect and a cognitive bias imputes an effect. or there’s some effect and the self-reporting messes up and reports no-effect. lot of stuff can go wrong.
or maybe “how do you explain the thing you cannot explain” is just a koan?
What does your model predict about the impact of precocious puberty, and how does that compare to the available data?
What does your model say about the impact of delaying puberty, and how does that compare to the data on adolescents who use puberty blockers and then undergo puberty at a later age than most?
When trans people medically swap out one of the standard sets of hormones for the other in adulthood, it’s often referred to as “second puberty”, because there’s a lot of similarity to what happens when the endogenous puberty hormones first show up. What does your model predict should cognitively result from doing puberty again, and how does that compare to the available data? (for this, definitely consider trans people who choose not to use hormones as a control group)
Your childhood home probably has greater total value to you than it would to a buyer. This particular house is important to you, and the odds of not losing a ton of money on replacing this particular house if you change your mind later are miniscule. Consider renting it, perhaps through an agency, for an amount that would allow you to rent something smaller in a safer location.
The exceptions which come to mind, where holding onto a house would be throwing money away long-term, are if it’s a manufactured in a trailer park with lot rent, or if it’s in an expensive HOA.
I think you probably shouldn’t rent to someone you know. “Move back into the property you own” is a good safety net for if your finances change suddenly and substantially. If you retain ownership, it decreases the probability that you will have no housing at all in many worst-case scenarios. Choosing not to renew the lease on a good tenant would be hard; declining to renew on a friend who was counting on you for their housing would be much harder.
It could be a good compromise to try to find your happy price—if someone offered you $1 for the house, you would laugh at them. If someone offered you $1B for the house, you would likely accept immediately. Somewhere in between those extremes is a price that would almost certainly get it sold fast; above that is a price that would be enough money to soothe the heartbreak of losing a childhood home. You could try listing the house at a price based solely on your feelings rather than market analysis, see whether it sells in a timely manner, and rent it out if it doesn’t.
If the structure turns out to be long-lived, keep an eye on that plastic sheeting, and pull it off as soon as you start to see it decaying from sun damage. It’s the kind of stuff that will fill a yard with hard-to-pick-up nastiness if left to the ravages of nature for too long.
If it’s the type of plastic designed for greenhouses, it’s likely to hold up much better. A scrap of pool liner or pond liner would make a good replacement. In an urban area, I’d keep an eye out for those corrugated plastic yard signs and staple them onto the roof like shingles.
There will be times when fear is telling you something correct and actionable, and should be heeded.
There will be many, many times when fear is telling you lies.
This is one of many opportunities to practice telling the difference.
I hope that math gives you a chance to see yourself overcome a personal challenge—observation shapes self-perception, and I think it’s better to perceive oneself as capable of overcoming hard things than otherwise.
The funny thing about numeracy is that it does double duty in the modern era: First, it scares away so many people that competition is reduced for certain jobs, and in turn, compensation tends to be higher for them.
Second and much more relevantly, slogging through calculus teaches you to notice certain categories of problem and form expectations about how their answers should look. You don’t have to keep on doing calculus once you’ve figured out how to use it, just like you don’t have to keep on taking driver’s ed class once you’ve gotten your license. But having a background in higher math—get through basic calculus and also give formal logic a shot; it’s a radically different flavor and can be much more palatable for some—can make it feel natural and obvious to stay on the correct side of compound interest at all times, which is critical for comfortably sustaining oneself in the current world. Money is math; financial instruments are math. Salary negotiation, like “should I push for higher base compensation or a bigger sign-on bonus or stock, and how does this depend on the type of stock”, is a form of math problem that tricks people into choosing lower over higher total compensation very frequently.
I hope that you can choose the path for yourself where you do not have to live in fear, nor wonder how much better things could have been if you’d just been a little braver. And if you want to commiserate about math, you’re welcome to mail me or whatever—it was my least favorite subject in school and I’ve failed many classes in it over the years, but the tech for finding interesting snippets of it and knitting them together into better general knowledge has improved vastly since I was in university.
Way to bury the lede there—you’re not “unsure”, you’re too easily daunted by mathematics.
You should stop running away from things that challenge you, and figure out how to enjoy the math that you’ve been getting stuck on. Look at it as a concrete example of the general life skill of improving yourself, if you like. It’s 2025; we are right this moment in the sweet spot of LLMs being cheaper than they’ll probably end up eventually (remember Uber and Lyft pricing back when they were new?), so you have a superintelligent tutor in your pocket that can do literally anything you can imagine to make maths more approachable.
Got a favorite fandom? I’ll bet any modern LLM can write decent fiction of it where your favorite characters learn and use the math you’re trying to wrap your head around. Prefer learning by examples? It’ll make you more examples than you’ve got time to read. Want to learn by listening to humans chat about it? notebooklm generates podcasts, or any of them can help direct you toward human-created content that matches your exact preferences. I don’t know what the fix will be for you, but I know that tackling your actual problem and figuring out how to pass classes you start off hating will improve the ecosystem of your overall capabilities far more than any major switch you could do right now.
If you solve your problem with math and still hate your major, then you can pivot into something else with the new life skill of working around problem classes in order to pursue your goals. If you don’t solve your problem with math now, you’ll see yourself being the kind of person who runs and hides when tasks get difficult which will reinforce that as part of your self-image, and you also probably won’t get as wealthy and successful as you would in the timelines where you prioritize the skill of numeracy.
I’ve never seen anything against that in public cryonics signup paperwork, but that would be a great question to ask one of the labs offering it!
The legal system would probably start caring a lot more about prohibiting it once we figure out how to get people back after cryonics.
I was with you right up to the part where I heard you implying that any large group of people can be effectively compelled to take any medication consistently over a long period of time. If you have some great way to get everyone to take all the drugs they’re supposed to, then pretty please apply it right now to causing people to take their full courses of prescribed antibiotics instead of just stopping once they “feel better”, and that’ll help slow down our accidental development of resistant bacteria.
Epileptics are a convenient example, because it’s so hard to imagine anyone insisting that they wouldn’t be themself without their seizures, and we don’t hear much about people deciding that the side effects of the medication for it are worse than the disease.
I suspect that the reality of (ongoing) medical treatments for criminal behavior is likely to more closely parallel the challenges that schizophrenics encounter with medication. In that case, the perceived subjective harms of taking the meds are much more closely balanced to the perceived subjective benefits. Most actual everyday average people, the kind of people with the deficits of luck or judgement that often correspond to incarceration, act as if they deeply believe that they should stop taking a medicine once they feel better from the condition it was prescribed to treat.
And even if people can be “fixed” in a one-off treatment, I’m getting popcorn for the lawsuits about religious discrimination when the system eventually encounters a prisoner of a faith with anti-medical views.
Mostly the benefit to me of doing the full stack of a project—design/cad/build, sketch/draft pattern/ sew garment—come from learning something new about the situation partway through the process, and being able to immediately re-open prior “closed” decisions from earlier in the process to take full advantage of what I realized partway through.
When I do a 3D printing project, my first test prints tell me a bunch of details about exactly how my particular printer handles this one particular part, and it’s low effort to assimilate those observations into changing how I design or orient or support it for future attempts. Or I can revisit the entire plan of printing a part and fabricate it in a different way instead. In sewing projects, my pattern design and mockups and the fabric I’ve picked all inform one another, and when I learn that the fabric would do particularly well or poorly for a given detail, I can immediately revisit the top-level plan and change the details to work better with the material.
This effect is more pronounced when I’m doing projects farther from my established skill set. I rarely get the benefits I’m talking about on projects where I’ve done it a bunch of times already and know exactly what will happen. Once I know exactly how something will go, it’s easy to outsource the process—“here, do this exact task on these exact things and it’ll definitely work”. But more often, I’m not yet at that level of expertise and familiarity, so I’m learning new things in the course of a project, and it’s beneficial to be able to apply the new insights wherever they’re most impactful. The friction of outsourcing a component, waiting on someone else to do it, etc opposes my process of applying newfound knowledge to the project’s entire top-level description.
The CAD design could be passed through a processor to generate the G-code (instructions the CAM processor reads and follows) and then some machinist would review and “fix” tool path or order to make the code more efficient for production.
This captures another angle on the question of whether one should learn a skill or outsource it: if the same person fixes the tool path and designs the component that needs the path fixes, that knowledge will inform their design choices on future parts. If there’s 2 ways you could draw a part and have it work how you need, then having the skills to fix the tool path and the knowledge to spot that one option would have worse path problems than the other will help you differentiate between the actual costs of the superficially interchangeable higher-level design options.
In my experience (beginner/intermediate CAD skills), this depends on how clearly you are able to explain the exact part you want.
If you only want parts which are easy to clearly describe, you can hire someone to CAD them up… or you could describe them to an LLM and have it emit files that you can convert to your CAD of choice.
If you want something less well-specified, or you want to “think out loud” and continually adjust your part’s design as you bring it into existence, the friction of having someone else running the software becomes prohibitive.
A similar question would be, how much benefit do you get from being able to type over having the money to hire someone who has typing skills and a lot of experience? Maybe you’d be fine with giving up that control over what words go onto the page, or maybe you’re using typing skills for something like driving a game or a text editor at a finer-grained level of control than would otherwise be available. Same with CAD.
For the pants, start with a pair that wash how you want them to, then take them to a tailor to adjust the shape to be correct. Then you have pants to wear while you spend months and hundreds on the skills and tools that it takes to compare with garment-factory quality on the hard parts like sewing through 4-6 layers of strong fabric.
Making pants nicely from scratch is a losing proposition on all axes except bragging rights, compared to adjusting an off-the-rack pair to fit as desired.
This reminds me that I recently learned about azolla—https://theazollafoundation.org/azollas-uses/for-food/. Individuals who want to do hands-on experimentation could grow some and explore culinary uses.
(If the human has a more than a million times as much moral status as a bacterium, which seems likely)
Apologies in advance if this sounds rude, I genuinely want to avoid guessing here: What qualifies the human for higher moral status, and how much of whatever-that-is does AI have? Are we into vibes territory for quantifying such things, or is there a specific definition of moral status that captures the “human life > bacterial life” intuition? Does it follow through the middle where we privilege pets and cattle over what they eat, but below ourselves?
Maybe I’m just not thinking hard enough about it, but at the moment, every rationale I can come up with for why humans are special breaks in one of 2 ways:
if we test for something too abstract, AI has more of it, or at least AI would score better on tests for it than we would, or
If we test for something too concrete (humans are special because we have the DNA we currently do! humans are special because we have the culture we currently do! etc) we exclude prospective distant descendants of ourselves (say, 100k years from now) whom we’d actually want to define as also morally privileged in the ways that we are.
I fear it would be a stupid mistake to pass on the opportunity to inquire: Where does one find the form?
I wonder what sources of information on clothing he believes that she is privy to and he is not.
He’s asking her for rules about clothing in a way that sounds like he’s asking her to choose a specific outfit.
She’s replying as if he asked whether a specific outfit was necessary, and as if she assumes he already knows the rules governing how to select an outfit for the occasion.
I wonder why such a smart guy has had the same unproductive conversation 100 times instead of seeking alternative ways to communicate the underlying question more clearly, such as “would <clothes I usually wear> be acceptable for this?”