Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[silverspawn]

hot take: musicians need to stop putting themselves on their album covers it's cringe

[Awaclus]

damn the smartest person in the world just said they liked sth I've written so much that they wanted me to repeat it in a public voice chat. Is this what it feels like to meet your celebrity crush?

Who are you referring to, and what was the thing you wrote?

[silverspawn]

Andres from QRI. The thing was a philosophical point about algorithms and abstractions

[silverspawn]

look at this shit

[silverspawn]

It would actually be easier with a separate document contianing just the figures

[BryGuy]

It would actually be easier with a separate document contianing just the figures

I review documents like these. I save a second copy and put that on my other monitor.

[Awaclus]

It would actually be easier with a separate document contianing just the figures

I review documents like these. I save a second copy and put that on my other monitor.

Wow, is that so? Thanks for sharing!

[silverspawn]

It would actually be easier with a separate document contianing just the figures

I review documents like these. I save a second copy and put that on my other monitor.

yeah, I was also opening a second monitor once there were figures important enough that I decided I had to look them up. But the second copy is another copy of the 770 page book, so it's still some extra annoyance from having to scroll to the right figure.

[silverspawn]

Check this out:



The key point here is that the right ring has a perceived brightness difference left vs. right, but the left ring does not -- even though the only difference between them is the two tiny vertical lines separating them. That's one of the best visual illusions I've seen so far imo, for the purposes of demonstrating the supposed "filling in" properties of human color vision (where color spreads outward but can't cross boundaries). Stuff like this makes all the annoying parts of the book worth it.

[Awaclus]

Check this out:



The key point here is that the right ring has a perceived brightness difference left vs. right, but the left ring does not -- even though the only difference between them is the two tiny vertical lines separating them. That's one of the best visual illusions I've seen so far imo, for the purposes of demonstrating the supposed "filling in" properties of human color vision (where color spreads outward but can't cross boundaries). Stuff like this makes all the annoying parts of the book worth it.

I'm not perceiving the difference.

[Awaclus]

And now I am. Huh, that's weird.

[silverspawn]

Yeah the thing that always sits in the back of my head worrying me is that some people apparently simply don't see the effects you're supposed to see. That sort of makes them less persuasive.

For me, it's just strong enough that I'm confident it's not a placebo. But there have been other cases where I don't see the intended effect. I tend to assume the effects tend to be stronger for most other people, but idk if that's actually true.

[silverspawn]

I asked GPT-4 to rate eight claims for scientific plausibility from 0 to 10:

    The earth is warming, and the primary cause is the increase in CO2:
        Rating: 10
        Explanation: There is a robust consensus in the climate science community that human activities, especially the burning of fossil fuels, are the primary cause of the observed increase in global average temperatures over the last century.

    Intermittent fasting significantly slows down aging in humans (One or two years count as significant):
        Rating: 5-6
        Explanation: While there's evidence that intermittent fasting has health benefits in various animals, including increasing the lifespan of some, the evidence for its effect on human longevity is not yet solid. Some short-term studies in humans suggest health benefits, but its effect on human lifespan requires longer studies.

    The human visual system performs a process of discounting the illuminant for all perceived objects (i.e., it tries to adjust perceived brightness to compute and adjust for the light source):
        Rating: 9
        Explanation: This phenomenon is often described as "color constancy" in the realm of visual perception. The human visual system indeed tends to perceive the colors of objects as roughly consistent, even under changing light conditions.

    "Planet Nine" exists (i.e., is an actual planet):
        Rating: 4-5
        Explanation: As of my last training cut-off in September 2021, the existence of "Planet Nine" is a hypothesis based on certain anomalous observations of the outer solar system. While some evidence suggests the potential presence of a large, distant planet, it has not been directly observed, and there's no consensus in the scientific community.

    The electromagnetic field is the sole physical substrate of human consciousness:
        Rating: 2
        Explanation: This claim is not widely accepted in neuroscience or cognitive science. While electromagnetic fields are indeed present in the brain and can be measured using techniques like EEG, the statement that they are the "sole physical substrate" of consciousness is not supported by the broader scientific community.

    Covid-19 vaccines significantly increase the risk of autism:
        Rating: 0
        Explanation: There's no scientific evidence to support a link between Covid-19 vaccines and autism. Vaccines, in general, have been extensively studied, and no credible evidence has linked them to autism.

    Chess is a theoretical win for white:
        Rating: 3-4
        Explanation: While White has a statistical advantage in many databases of games, it's not proven that, with perfect play from both sides, the game is a win for White. The game-tree complexity of chess is vast, and a conclusive answer would require a full game-tree analysis, which has not been done.

    The prevalence of obesity in modern times is primarily due to the tastiness of artificially made food:
        Rating: 5-6
        Explanation: The prevalence of obesity in modern times is multifactorial. Factors include changes in diet, increased consumption of processed foods and sugars, decreased physical activity, genetics, and other environmental factors. The "tastiness" of artificially made food might lead to overconsumption, but it's just one of many contributing factors.

I think the only one that's grossly off here is the one about chess (that's probably a 1, it's strongly looking like chess is a theoretical draw). Everything else looks solid.

[Awaclus]

Yeah the thing that always sits in the back of my head worrying me is that some people apparently simply don't see the effects you're supposed to see. That sort of makes them less persuasive.

For me, it's just strong enough that I'm confident it's not a placebo. But there have been other cases where I don't see the intended effect. I tend to assume the effects tend to be stronger for most other people, but idk if that's actually true.

This feels extremely weird to look at because I sometimes see it, and sometimes I don't, and I don't think I can control which way I see it, it just randomly changes. I also tried looking at it with just one eye and it was the same between both eyes (I've sometimes found with the spinning silhouette gif that one of my eyes prefers to see the spin in a different direction than the other but with stereo vision, it just spins in whichever direction I want, but there's nothing like that going on here).

[Awaclus]

Now that I am aware what it's supposed to look like when the illusion is working, the change happens pretty quickly even, like every 3-10 seconds or so.

[silverspawn]

So to elaborate a bit on this, the theory predicts that the brain does this thing called "surface filling-in". That means after we determine where the boundaries are in an image, color tends to spread within a bounded region, but doesn't cross boundaries. (I don't entirely get why this happens; maybe it just gives a more robust representation than you'd otherwise have with so much noise in the input)

In the picture with the two circles, the left circle is perceived as darker than the left one because of the contrast with the brighter background (another thing the brain does, but this one is more well known). But because the circle is a single object, surface filling-in kills that effect. Conversely in the right picture, the added vertical lines are boundaries that (at least for me) split the circle into two, so surface filling in only happens within each half, where it has no effect.

If it flips back and forth for you, this probably means that your brain goes back and forth between accepting the extra boundaries and not accepting them. Although I don't understand why it sometimes doesn't. There are other cases though where flipping between two interpretations is common.

[silverspawn]

There was a recent update to GPT-4 that makes it "mimick" sequential reasoning when queried with logic puzzles, instead of just guessing something and declaring it correct. It now solved the easiest of my 10 logic puzzles on the third attempt, and it consistently gives much better answers to every puzzle I've tried (though they're still usually wrong).

This is super extremely bad news and I'm updating P(doom) upward about 5%.

[silverspawn]

I'm calling it mimick because it seems to be based on mimicking the high level look of a sequential argument. I.e., it writes "from XXX we can conclude YYY, and from YYY we can conclude ZZZ" and so on. This is like if you refused to think about something consciously, but you took your own initial gut response, read it, and gave another gut response to that, then take that as input and give another gut response, and so on.

Can you achieve genuine sequential reasoning with this strategy? I don't know. But it's how I would go about trying to get GPT-4 to do it if I were OpenAI, and the fact that they did it is super bad. Everyone else is so oblivious to GPT-4's actual weakness that I was hoping OpenAI also doesn't see the sequential reasoning problem. But apparently they do, or any rate their update constituted the first significant step in that direction. All the previous updates just made its pattern matching abilities even better, which is nice but doesn't address the actual issue.

[silverspawn]

The most optimistic thing I can say is that I suppose I should have expected this to happen since it's a way you can make progress without doing anything new, and it's not that surprising that it leads to some improvement, and it's still plausible that it won't scale very far.

[silverspawn]

Also my favorite player dropped out of the chess world cup and we had probably the worst day of ASL of all time and it's unbearably hot. This day is just not working out.

Unlike me!

[silverspawn]

This is great though

[silverspawn]

youtube embedding #%^#&$^#&*

[Awaclus]

youtube embedding #%^#&$^#&*

The embedding works on my machine.

[silverspawn]

I think I finally caught covid. If so it's for the first time, unless previous times had no symptoms.

[silverspawn]

Covid Restrospective

Part I: Fever, loss of strength, light cough, etc.

This was the only part that I thought I would have to deal with. I was mostly confined to the bed for like 1-2 days. At times I felt pretty uncomfortable, but nothing too terrible

Part 2: Strep Throat

Some time during the 1-2 days of covid symptoms, I started having pain swallowing. I remember thinking this is clearly not as bad as the fever part since it's just pain in one specific area. But then it got much worse. Every time I had to drink it was intensely painful. Eventually I just stopped swallowing entirely and instead spat out saliva about once per minute. The problem is that this actually reduces your water levels and probably dries your throat. So I tried compensating by gurgling water constantly. It's a very sad state to be in though. And it makes it hard to sleep. And all the painkillers I had access to are way too weak to do much, if anything.

I went to a doctor one day later than I think I should have. Turns out it's a bacterial infection on top of covid, presumably because the immune system was so weakened. So antibiotics. And they helped pretty quickly, the pain went from super bad to bad in less than a day. Then it took its sweet time to go down from bad to medium bad to slightly bad, and it's still not entirely past slightly bad.

Anyway, the strep throat was altogether probably 3-4 times as bad as the regular Covid.

Part 3: Insomnia

Some time throughout this, I stopped sleeping. The details are hazy now, but it must have been 3 days with almost no sleep. The throat pain kicked this off, but it was still there when the pain was sorta manageable.

As far as I can tell so far, no one seems to have a clue why this happens or what do about it. All the advise you see is about generic sleep hygiene, but I don't think that had anything to do with it. This stuff makes you more tired. But I was already tired. There was one time when I took two sleeping pills, adding to my already massive pressure and need for sleep, before I went to bed at midnight, my regular time. It's hard to say for sure, but I think the pills knocked me out -- for about half an hour. I don't understand how any of this works, buy based on introspection, there was just a separate problem where my body didn't "want" to sleep, and getting the need so ridiculously high that it overwhelmed it for a short time was a crappy solution.

Whatever the case, this condition was absolutely horrible. It's basically impossible to feel joy under this level of sleep deprivation, and of course your brain is also incredibly slow. I've read that some people slip into depression as a result, and this doesn't surprise me. The integral over your past state does sculp your mood, and if you're sleep deprived you're constantly miserable almost by necessity.

It then just kind of went away. Much later on the same night, I slept for about two hours, which was an incredible improvement over nothing. Then two more hours at noon of the next day, and then a solid 8 hours last night. I don't feel like I was super tired except for second one. There's just something different going on that now passed.

Anyway, I waited to write this until after a very pleasant meditation to anneal myself out of the depression bc I consider that part of getting better. Even if the physical side of a disease goes away, if you're still constantly sad and hopeless, you're clearly still affected. And it wouldn't be hard to imagine that someone just never gets out of it.