jawarner

Interesting, in my experience Astra is a marked improvement over both Sol 5.6 and Fable 5.1. Its output feels a lot more natural, and it is just less "dumb." But individual experiences may vary.

Noting the theory papers on Kolmorogov complexity. For those not familiar, Ilya argues that the reason why neural networks generalize -- why they work at all -- is because they are effectively finding a simple description of their training data, converging down onto the limit of the Kolmorogov complexity. [1]

[1] https://www.youtube.com/watch?v=AKMuA_TVz3A

For those who don't know -- while Zachtronics is no longer making games, Zach Barth is still active now under the company Coincidence Games. They just game out with a spacecraft engineering puzzle game:

https://store.steampowered.com/app/2536720/UVS_Nirmana/?cura....

Looks like this may have received the HN Hug of Death. I'm getting "Too Many Requests" error trying to load the images.

Is this AI? You've copied the best comment and put into AI-speak.

Mawson et al. 2017 (two papers) – internet survey of homeschoolers recruited from anti-vaccine groups; non-random, self-reported, unverified health outcomes. Retracted by the publisher after criticism.

Hooker & Miller 2020/2021 – analysis of “control group” data also from self-selected surveys; same methodological problems.

Lyons-Weiler & Thomas 2020, 2022 – data from a single pediatric practice run by one of the authors; serious selection bias.

Joy Garner / NVKP surveys – activist-run online surveys with no verification.

Enriquez et al. 2005 – a small cross-sectional study about allergy self-reports, not about overall neurodevelopment.

Large, well-controlled population studies (Denmark, Finland, the U.S. Vaccine Safety Datalink, etc.) comparing vaccinated vs. unvaccinated children show no increase in autism, neurodevelopmental disorders, or overall morbidity attributable to recommended vaccines.

Take a look at the next sentence from your linked article.

“I am pro-life personaly [sic] but it wasn’t those,” he said, using the jail’s internal messaging system. “I will just say there is a lot of information that will come out in future that people will look at and judge for themselves that goes back 24 months before the 14th. If the gov ever let’s [sic] it get out.”

I think for this to be interesting to an outside observer, it should at least address the bulk of the debate around the topic. If the GP is right and it's missing the key points of the anti-abudance critique, then I'm afraid it's missing the forest for the trees and misleading to a general audience.

Insofar as "the problem" is sifting through pages and pages of documentation to find the relevant information for an answer, I think that current-day LLMs are actually quite capable at this.

You can measure the sharpness of the position, as in this paper section 2.3 "Complexity of a position". They find their metric correlates with human performance.

https://en.chessbase.com/news/2006/world_champions2006.pdf

Real world systems are complicated. In theory, you could do belief propagation to update your beliefs through the whole network, if your brain worked something like a Bayesian network.

Isn't that Edwin T. Jaynes example just p-hacking? If only 1 out of 100 experiments produces a statistically significant result, and you only report the one, I would intuitively consider that evidence to be worth less. Can someone more versed in Bayesian statistics better explain the example?

Yes, the possible programs are enumerable, and you can start searching with the least complex programs and work your way up in complexity. Once you find a program that explains the available data, you cannot guarantee it will continue to explain possible future data, unless, like you mention, you constrain the program space to a finite set. What you're describing is generally how people make models of the external world.

I don’t know if it’s fair to say that revolutionary wealth redistribution movements, like Occupy, are failing because everybody is distracted with social justice movements. The Left is a coalition for social justice and also for wealth redistribution, to varying degrees. That’s why I’m saying that I’m not seeing a direct connection where a success in one area subtracts from progress in another.

If you're just saying that rainbow capitalism is in alignment with capitalists' interests, and anti-capitalism movements are not, then that makes sense.

Isn’t that because of the BLM movement? I’m not seeing how any of these events are connected.

No, the text is AI too. It is crazy.

Just for the sake of argument, I can name some an entire field of science that was invalidated in light of genetic & neuroscience evidence: phrenology. At the time, it was the newest advancement in the gleaming era of scientific Enlightenment. It just happened to justify colonial policies of that time. Now, a couple hundred years later, we're walking back on a widely supported but misguided "scientific" field.

Are those two events connected? They both happened in the same decade and received some press. That's all I can come up with. And both things you mention -- LGBT rights and class consciousness -- have been around for a while.

Yeah but your enemy is also moving forward because they’re trailing you.

If you think about it, those lives actually were saved. People either didn't play the video game, or they spent a moment in the loading screen in quiet contemplation.

The tweet is in response to a preliminary paper [1] [2] studying text found in images generated by, e.g., "Two whales talking about food, with subtitles." DALL-E doesn't generate meaningful text strings in the images, but if you feed the gibberish text it produces -- "Wa ch zod ahaakes rea." back into the system as a prompt, you would get semantically meaningful images, e.g., pictures of fish and shrimp.

[1] https://giannisdaras.github.io/publications/Discovering_the_...

[2] https://twitter.com/giannis_daras/status/1531693093040230402

The reaction diffusion / NCA models have been applied to videos recently — check it out:

https://wandb.ai/johnowhitaker/nca/reports/Fun-with-Neural-C...

Not a language model, but there was this classifier mapping pig squeals to contextual information about the animal [1].

For human language models there are obvious practical uses and we can intuitively evaluate the quality of the models. For animal communication you need to use other biological variables, like valence in the pig communication example.

I do remember one where they generated artificial frog mating calls and were able to generate artificial ones that were exceptionally attractive to female frogs, but that was from a while back, and I can’t find the link.

[1] https://www.nature.com/articles/s41598-022-07174-8

There are physiological adaptations as well. People in hotter climates are still going to get good sleep. It’s just when there are hotter temperatures than people are used to that sleep is impaired, and also that adaptive mechanism only works so well, and there is disparate impact for people in different climates. At a certain point, there are heat waves or very hot temperatures that simply aren’t conducive to human life.

100% agree about the urgency to make an economic system that is compatible with climate stability and thus with human life.

The fact that heat stress affects different populations with such different impacts is super important to climate policy and diplomacy and geopolitics. The authors in this case identified a concrete, statistically sound demonstration of disparate impact of heat stress.

That’s something I wish I had emphasized in my top-level post.

If you don’t mind I made a clarifying post:

https://news.ycombinator.com/item?id=31470654

Their angle was health equity in the context of global warming and how it impacts different populations to various degrees. That’s a common trend in global warming related issues. Their argument is fair enough.

My issue with the paper was specifically with the projections they made, which extrapolated the effect on sleep caused by the weather all the way to effects on a 50-year interval caused by warming trends. But by the authors own admission people can physiologically and technologically adapt — at least partially — over long enough time periods. To be frank I think it’s a way to make a compelling headline. They have the statistical tools they have, it would be incredibly hard to account for long-term adaptation, and so they come up with some statistical estimate of the long-term trend. I see why they did it, but from a readers perspective it’s okay to be critical.

The gist: On hotter nights people get less sleep, this being especially the case for people who are poor (limited access to AC?) and who already live in hot climates (100->110 deg is more noticeable than 70->80).

It’s nice data. They gave sleep tracking watches to 47,000 subjects for a few years, and this is what came out of it.

Their climate change angle is suspect. It probably helps publish to be relevant to a real-world problem. But their final paragraph undermines the projections they try to make: people adapt to the long-term weather patterns, and they’ll likely partially adapt to climate change occurring over the course of 50 years. Of course it’s still relevant from a health equity standpoint to consider.

You ever toss and turn at night because it’s just too hot to sleep? Someone moving from the North to California would likely experience that. I know I did when I moved to Texas and my air conditioning unit broke down.

The authors say there is adaptation to the local climate so someone living near the equator might still get good sleep. The data is based on weather, and it checks out; it’s hard to sleep on a hot summer night.