espoire

The random seed for watermarking doesn’t use the whole context, just a few tokens immediately preceding each choice. This is deliberate; it enables detection of e.g a couple of AI-written sentences taken out of context.

If you added a per-user private key into the random seed, you could indeed check whether a specific user used the same AI to write a block of text.

…but asking which user (if any) used that AI to write the text would be quite expensive. The only way to test that would be to rerun the watermark test for every individual user, and you’d probably have a few false positives in the mix just by sheer numbers. A one-in-a-million fluke should happen about once per million users, after all.

Sharing a thought trace I had after reading the above, in the hope that my thoughts may entertain someone. I feel a little dumb for having had them.

Wait, how can you be sure that set is distinct from just “all the primes”?

Oh, wait. Duh. Graham’s Number is very very divisible by 3.

On the hardest climb of my life so far, half my group turned back at around 85% of the way up. This was a correct decision, and things would have turned scary if they hadn’t. Indeed, the decisions to turn back are, for each of the people involved, the single thing they’ve ever done that I most respect.

The leader and organizer of our expedition was also the least physically-fit. Around mid-afternoon, he was feeling exhausted, and his winter coat was soaked in his sweat. It was late winter, and while the ambient temperature had been quite tolerable during th... (read more)

As someone who has successfully scaled two (very very easy, by comparison to those mentioned in this post) mountains: confirmed. It has been worthwhile in retrospect. My average happiness has been noticeably improved since I climbed the first one, which given what I’ve read in psychology is a big claim (allegedly hardly anything ever enduringly alters average happiness).

That said, I climbed much much easier mountains with almost no danger, summit and back down all within one day’s light (well, almost; it went dawn to late twilight at the end of the longer climb), no specialized equipment needed beyond a fancy backpack full of water plus one sandwich. My climbs were also in-the-moment fun.

It would be a grave error to assume that these effects remain additive, without side-effects, when you combine them all.

It’s my (weakly held) belief that some of the human intelligence genes act through lowering the happiness set point. Unhappy more often —> more problems to solve —> more practice solving problems —> better at solving problems. If true, even for a smallish subset of the human intelligence genes, then stacking that would be both horrifying and seems intuitively to me like it would fail past a certain point.

Evidence: ~all the cognit... (read more)

I can consciously move my eyes smoothly. It was a learned ability for me, starting and practicing from various related starting operations like “tracking an imaginary moving object” or “crossing ones eyes”.

It’s particularly obnoxious with traffic lights in the USA (at least, in Wisconsin, Illinois, Connecticut, New York, and Texas). Unless I actually point my head directly at the traffic lights, the actual light appears outside the signal, superimposed on e.g. the sky. Which can make them very hard to see if the sky is particularly bright or similar color. Stupid monochromatic traffic signal light…

Direct forced stabilization of an image on the retina using specialized devices also causes images to fade.

I’ve always had conscious control of my “automatic” eye movements, in that it takes a very small continual mental effort to enable those movements, and if I mentally relax the movements stop. (And with a deliberate mental effort, I can substantially amplify the random movements, which other people describe as very unnerving to see my eyes doing. I can move my eyes independently of each other like a gecko, too, a skill I developed via deliberate practi... (read more)

In fact, the photoreceptors of the Human eye aresomewhat sensitive to near-ultraviolet, but the natural lens screens these out to protect the retina. Older artificial lens replacements sometimes didn’t, and individuals with such a lens (and the rare individual without a lens) sometimes reported seeing near-ultraviolet as a blue-white or purplish color.

Neat!

Also striking in that I see an extra color past the violet end of a natural sun-lit rainbow (as compared to e.g. photographs or artistic renderings of a rainbow, which omit the last color) — and yep, blu... (read more)

Actually, that’s an interesting thought. The usefulness of the notion The Tao of Physics implies EITHER that there actually is some underlying distribution from which possible physical laws were sampled for our universe, OR (perhaps as a subset of the first horn of this dilemma?) that there is an intelligent designer with actual preferences over mathematical style of physical laws. The only non-infinitesimal scrap of evidence we have for a supposed creator with any preferences at all beyond our universe’s mere existence.

I was surprised to hear animal skins called a poor protein source, though on reflection I guess it fits my priors that you wouldn’t want to eat all your protein as skin.

The reason this surprised me is that, in my area and social circles, a lot of folks eat ~zero skin and noticeably benefit from adding some to their diet, e.g. as one cut of chicken a week, eaten along with the skin that would be on that if not removed. I’ve been going around informing people “hey, you should eat more skin”, when what I meant was “you should eat more [than zero] skin”.

Aside: I misunderstood “accessible“ above for quite a while. I thought “this platform is not accessible” meant “there’s been a road closure or a structural collapse or something, which is preventing anybody from getting in or out”.

At the risk of sounding gauche, I really do think general purpose communication ought to spell out “handicap accessible”.

I would and have totally wrote essays pushing boundaries on what an essay is

When I was a youthful hellion, I handed in “Stupidity is Flammable: On the Use of Excerpts, Artful Verbiage, and Rhetoric to Disguise a Lack of Content”, a whole-assed high-effort essay about how to refuse to do the research and then submit a bogus essay.

I remember it getting a ‘C’ with no specific feedback, and being offended at the time.

ignores things like how quickly different materials are shed from the body

Seems like a reasonable thing to do, given which substances we’re discussing. Lead‘s half-life for leaving the body is several decades — it’s a fair approximation to say that it does not leave the body.

I notice in the data above that, where multiple chocolates from the same brand are available, the darker (higher percentage) variant tends to have less of both lead and cadmium, and they’re different by a fairly large amount.

Going only off that observation: not the cocoa solids. Further suspicion: milk chocolate might be on average worse?

I’ve had really surprising success with this method in past. Admittedly many issues have stumped me, but in three cases I managed to trace the causes back to worse-than-useless dietary advice, causing the person to give themselves a nutrient deficiency by avoiding some important class of food, which then caused a whole host of symptoms. Telling them to eat the necessary food brought immediate and unprecedented relief — at least during weeks where they actually ate any. As it turns out, changing one’s dietary habits is very difficult, and not just for weigh... (read more)

I agree with the suggestion for use of plainer language. Make the case for the astonishing significance within the text, but be as plain as possible in the terminology, else I fear you’ll lose the vast majority of your potential future audience.

I consider myself fairly erudite, and a connoisseur of etymology, and yet the phrase “effective chrono-lucropia” is very nearly opaque to me. I still don’t understand what the ‘effective’ part is meant to signify, and my best guess at dissecting “chrono-lucropia” is: chrono=time, lucre=wealth or money, opia=seeing. ... (read more)

I think it was intended as a caution against a common bit of anti-epistemology that maybe you neither have done nor have yet encountered from others.

Namely: constructing a contrived definition for some purpose, and then when people complain about the problems your strange definition is causing for them, the one defends their choices with “it’s just a definition bro, chill”, essentially.

Often this happens when some really deeply technical definition collides with the same word having escaped into common usage. …or as often prompts vitriol in social media: s... (read more)

Random thought: I think we humans named “negative“ and “positive“ curvature backwards. Positive curvature decreases the distance between objects as they seem to move apart. Negative curvature increases the distance between objects as they seem to move apart. I feel like a hypothetical world where the names “negative“ and “positive“ had been swapped at inception would be more human-intuitive.