SneerClub

cross-posted from: https://lemmy.world/post/11178564

Scientists Train AI to Be Evil, Find They Can't Reverse It::How hard would it be to train an AI model to be secretly evil? As it turns out, according to Anthropic researchers, not very.

edited to add tl;dr: Siskind seems ticked off because recent papers on the genetics of schizophrenia are increasingly pointing out that at current miniscule levels of prevalence, even with the commonly accepted 80% heritability, actually developing the disorder is all but impossible unless at least some of the environmental factors are also in play. This is understandably very worrisome, since it indicates that even high heritability issues might be solvable without immediately employing eugenics.

Also notable because I don't think it's very often that eugenics grievances breach the surface in such an obvious way in a public siskind post, including the claim that the whole thing is just HBD denialists spreading FUD:

People really hate the finding that most diseases are substantially (often primarily) genetic. There’s a whole toolbox that people in denial about this use to sow doubt. Usually it involves misunderstanding polygenicity/omnigenicity, or confusing GWAS’ current inability to detect a gene with the gene not existing. I hope most people are already wise to these tactics.

"glowfic" apparently. written in a roleplay forum format.

This is not a story for kids, even less so than HPMOR. There is romance, there is sex, there are deliberately bad kink practices whose explicit purpose is to get people to actually hurt somebody else so that they'll end up damned to Hell, and also there's math.

start here. or don't, of course.

I did fake Bayesian math with some plausible numbers, and found that if I started out believing there was a 20% per decade chance of a lab leak pandemic, then if COVID was proven to be a lab leak, I should update to 27.5%, and if COVID was proven not to be a lab leak, I should stay around 19-20%

This is so confusing: why bother doing "fake" math? How does he justify these numbers? Let's look at the footnote:

Assume that before COVID, you were considering two theories:

1. Lab Leaks Common: There is a 33% chance of a lab-leak-caused pandemic per decade.
2. Lab Leaks Rare: There is a 10% chance of a lab-leak-caused pandemic per decade.

And suppose before COVID you were 50-50 about which of these were true. If your first decade of observations includes a lab-leak-caused pandemic, you should update your probability over theories to 76-24, which changes your overall probability of pandemic per decade from 21% to 27.5%.

Oh, he doesn't, he just made the numbers up! "I don't have actual evidence to support my claims, so I'll just make up data and call myself a 'good Bayesian' to look smart." Seriously, how could a reasonable person have been expected to be concerned about lab leaks before COVID? It simply wasn't something in the public consciousness. This looks like some serious hindsight bias to me.

I don’t entirely accept this argument - I think whether or not it was a lab leak matters in order to convince stupid people, who don’t know how to use probabilities and don’t believe anything can go wrong until it’s gone wrong before. But in a world without stupid people, no, it wouldn’t matter.

Ah, no need to make the numbers make sense, because stupid people wouldn't understand the argument anyway. Quite literally: "To be fair, you have to have a really high IQ to understand my shitty blog posts. The Bayesian math is is extremely subtle..." And, convince stupid people of what, exactly? He doesn't say, so what was the point of all the fake probabilities? What a prick.

The Techno-Optimist Manifesto By Marc Andreesson

Eliezer Yudkowsky @ESYudkowsky If you're not worried about the utter extinction of humanity, consider this scarier prospect: An AI reads the entire legal code -- which no human can know or obey -- and threatens to enforce it, via police reports and lawsuits, against anyone who doesn't comply with its orders. Jan 3, 2024 · 7:29 PM UTC

Pass the popcorn, please.

(nitter link)

found at https://twitter.com/tracewoodgrains/status/1741544491490713853

the only copy of the suit I can find so far is here https://trellis.law/doc/163172995/brief-memorandum-points-authorities-filed

I'm called a Nazi because I happily am proud of white culture. But every day I think fondly of the brown king Cyrus the Great who invented the first ever empire, and the Japanese icon Murasaki Shikibu who wrote the first novel ever. What if humans just loved each other? History teaches us that we have all been, and always will be - great

read the whole thread, her responses are even worse

Is uh, anyone else watching? This dude (chaos) was/is friends with Brent Dill.

an entirely vibes-based literary treatment of an amateur philosophy scary campfire story, continuing in the comments

... while at the same time not really worth worrying about so we should be concentrating on unnamed alleged mid term risks.

EY tweets are probably the lowest effort sneerclub content possible but the birdsite threw this to my face this morning so it's only fair you suffer too. Transcript follows:

Andrew Ng wrote:

In AI, the ratio of attention on hypothetical, future, forms of harm to actual, current, realized forms of harm seems out of whack.

Many of the hypothetical forms of harm, like AI "taking over", are based on highly questionable hypotheses about what technology that does not currently exist might do.

Every field should examine both future and current problems. But is there any other engineering discipline where this much attention is on hypothetical problems rather than actual problems?

EY replied:

I think when the near-term harm is massive numbers of young men and women dropping out of the human dating market, and the mid-term harm is the utter extermination of humanity, it makes sense to focus on policies motivated by preventing mid-term harm, if there's even a trade-off.

I somehow missed this one until now. Apparently it was once mentioned in the comments on the old sneerclub but I don't think it got a proper post, and I think it deserves one.

From Sam Altman's blog, pre-OpenAI

Epistemic status: Speculation. An unholy union of evo psych, introspection, random stuff I happen to observe & hear about, and thinking. Done on a highly charged topic. Caveat emptor!

oh boy

archive: https://archive.is/uOP4y

When I click a link to LessWrong from this board, I receive a malware alert from my home gateway (Netgear Armor). Apparently it's their AI text-to-speech bot.

Question - any concerns about this? Google isn't helping me much.

URL is https: // embed.type3.audio/

Searching their site tells me that this is literally a feature and not a bug.

https://www.lesswrong.com/posts/b9oockXDs2xMdYp66/announcement-ai-narrations-available-for-all-new-lesswrong

TYPE III AUDIO is running an experiment with the LessWrong team to provide automatic AI narrations on all new posts. All new LessWrong posts will be available as AI narrations (for the next few weeks).

You might have noticed the same feature recently on the EA Forum, where it is now an ongoing feature. Users there have provided excellent feedback and suggestions so far, and your feedback on this pilot will allow further improvements.