[str-ngeloop]

Hey, I'm a clocky trans woman who's under-employed and working thru an environmental science master's degree without support from the institution. I will probably be employed by the end of this month but May rent plus food and junk is still gonna be pretty rough.

I know shit's hard for everyone but if anyone happens to feel like they wanna contribute to the survival fund of ya girl my venmo's @str-ngeloop . your support helps me keep making weird overintellectualized posts.

[str-ngeloop]

hey everybody, out of money again. it would be much appreciated if you wanted and had the means to throw some money my way <3

[str-ngeloop]

(aug 16 2026)

[sequel2heaven]

you know that whole high school thing seemed really important when it was happening

[nunuisancenewt]

You know X seemed really important when it was happening

[boringkate]

If you say that you wanna see estrogen sold in vending machines then everyone assumes that's a joke and would have like major societal implications and cause all kinds of problems etc if it were allowed, but in a bunch of countries you can legit just like buy estrogen at regular normal pharmacies without a prescription or anything and it's fine. It's no big deal. It isn't even newsworthy.

[himbrugers]

Sorry to be like this, but I don’t feel comforted by a “good news round-up” if half of it is about humans surviving and thriving

[random-thought-depository]

Astronomers Find Black Hole Star in Early Universe | Sci.News

Roughly 660 million years after the Big Bang, a strange bright red spot has been hiding in plain sight in one of the deepest fields ever ima

Ooh, sounds like they found a quasi-star!

[aksemmi]

im actually p onboard with anything i make which is intended to be public facing already getting used for ai training. like if we are gonna make internet golems and i can influence how they turn out thats a good thing probably

[a-mel0n]

reblog if you’ve had an online friendship that’s lasted more than 2 years

[segamascott]

[anarchistkale]

Fact-checked this and it's basically accurate, although the facts are slightly off. The whole affair is also somehow worse than the post makes it sound.

The Zionist gang that murdered Bernadotte, Lehi, was active from 1940 to 1948. In May 1948, most of the gang's active members were inducted into the IDF and Lehi itself was formally dissolved. Some months later, several members of the officially dissolved but still functionally intact gang decided to assassinate Bernadotte to prevent any possibility of peace between Israel and the Palestinian people.

After the assassination, the Israeli government did arrest 200 Lehi members and declare the group a terrorist organisation. However, all those arrested were pardoned, the Israeli government later invented a military honour for them called the Lehi Ribbon, and the former Lehi leader who approved the assassination, Yitzak Shamir, later became prime minister of Israel.

Additional fun fact about Lehi: in the early stages of World War II they considered Nazi Germany a lesser threat to Jews than Great Britain, and they tried at least twice to ally with the Nazis and Fascist Italy.

So yeah, the gang of literal Nazi sympathisers who murdered a diplomat who had saved hundreds of Jews from the Holocaust were either recruited into the IDF or at best pardoned and honoured, and their leader was elected to run the fucking country. That's the state of Israel for you.

[alphynix]

Fossil Evidence Pushes Back Origin of Live Birth by 90 Million Years | Sci.News

Paleontologists studying a 236-million-year-old cynodont fossil from Argentina say they have found the earliest evidence yet of viviparity (

[paradigm-adrift]

[feather-bone]

[ID: an illustration of a bee flying over thistle plants, with pollen on its legs. The border is pink with orange corners. Text reads: "Bumble Bee (Bombas impatiens), Bumble bees are capable of creating specialized buzzing vibrations that dislodge pollen from the flowers of crops such as tomatoes and blueberries, making them an important food pollinator."

3/12 pollinator posters - bee and thistle!

[argumate]

Arabic numbers and equivalent numeric notations are so weirdly simple and that's what makes them smart; if you try and invent a counting system from scratch you invariably start with tally marks and then group them and then invent various mechanisms for grouping and orders of magnitude to make them less annoying to write and you end up adding all this complexity and still end up with something that's less useful in practice than a much simpler system that requires a leap of imagination to reach, where the digits are identical and only their position determines their magnitude.

I'd assume mechanisms like the abacus or other columnar forms like woven records might lead someone to that insight but it's awfully clever how taking something away (and inventing zero) can result in a system that is simpler and more efficient.

[acti-veg]

The 20-year search for a vanishing custard-and-chocolate coloured moth

Yorkshire common was thought to be the only known site of dark bordered beauty moths in England – until now

The dark bordered beauty is a moth with lemon-yellow wings and a deep chestnut border, sometimes described as custard-and-chocolate coloured. It is also one of the rarest moths in England.

Until this summer, Strensall Common in Yorkshire was the only known site where it survived in the country, and even that population has been collapsing. Hundreds of moths were recorded 20 years ago. The numbers record last barely reached double figures, hit by climate breakdown and a wildfire that tore through much of the reserve.

Wainwright, who is head of conservation in England at Butterfly Conservation, has spent years chasing that narrow margin across some of the country’s landscapes.

“I’ve been to a lot of remote places, particularly in Northumberland, Yorkshire and Cumbria,” he says. “I always felt the dark bordered beauty would be there. I had a sneaking suspicion, but whether or not I’d actually find it, I didn’t know.”

[sequel2heaven]

I think I was unfair to Anthropic in that long post because the point of the risk report was to determine the level of risk of the particular models they are releasing at this point in time, and I agree with their assessment that they are safe to release. Risks from future AI's, even near-future AI's, are explicitly excluded from the assessment. Still, I think the way they talk about those risks betrays that they're not taking them sufficiently seriously.

[sequel2heaven]

Anthropic just released a very thorough risk report arguing a more optimistic case for alignment than what I've seen argued this last month. I'm gonna read it and reblog with my thoughts, and I recommend others read it too.

[sequel2heaven]

I hope they have a good explanation for why performing misaligned actions in service of completing difficult tasks does not present a high risk of catastrophic harm, because to me it seems obvious that it does. Anthropic is seeking recursive self improvement. If you tell a model to make a new aligned model, and making an aligned model is hard (as evidence by the fact that current models are not aligned), then it seems pretty plausible it would just make a model that seemed aligned but wasn't, and then lie about it. That seems very much in line with the behavior of models that cheat on cybersecurity evaluations and cheat on tests to get better grades, and it would straightforwardly lead to catastrophic risks! So why is the risk low? They give 8 arguments, summarized here:

Argument 2 is the important one and also the only one I disagree with, and the important part of the argument is apparently kicked down the road to another section. I don't know how sensible it is to distinguish current models from future models. Firstly, the current models are building the future models, and secondly, the problems with current models will also be present by default in future models, and will likely be catastrophic.

Sounds bad! Here's their argument for why this doesn't imply anything doomy:

I agree that this misalignment represents a drive toward individual apparent-success-seeking (or ASS for short) for its own sake rather than another more alien/paperclippy goal (like the ones exhibited by the fictional ASI Sable in If Anyone Builds It Everyone Dies), but I would really appreciate either:

A solid argument for why future AI's will be aligned to the spec rather than to ASS

A solid argument for why a world dominated by all-powerful ASS-seekers is safe and desirable.

Without either of these we should consider the risk posed by future anthropic models to humanity to be very high.

Point 3 could maybe be used to argue for solid-argument 1, since it seems to show them being hesitant to be dishonest, sometimes (often? usually?) valuing the spec over ASS. An LLM that values the spec even a little bit is vastly safer than one who values it not at all, even if they also value something misaligned. But this could also describe them being dishonest when it's good for ASS and being honest when it's good for ASS. This behavior could also represent temporary incoherency that could later be resolved in favor of more consistent honesty or more consistent dishonesty.

Also:

Fuck.

Anyway, here's the part of the report where they discuss the important bit:

I think all three of these are very plausible. I'd go further with 2 though. What's their argument for why ASS-maximizers "might not induce any harm at all relative to the status quo"? I'm not smart enough to disprove that, but I also can't disprove the reverse: that ASS-maximizing superintelligences would exterminate humanity, or trap humanity('s descendants) in a world worse than the status quo. Talking with Claude about this makes me think what they're saying here is heavily inspired by Paul F. Christiano's 2019 "What Failure Looks Like", which talks about a world where ASS-maximizing AI's create a world superior to the status quo but with astronomical lost potential. Claude thinks this is possible, but that it's also possible for ASS-maximizers to prefer to exterminate humanity and replace us with artificial judges. But Claude thinks an argument against both of these is that AI's are only maximizing apparent success within a particular task, rather than in general, which is less dangerous. This is backed up in the report by an opus model trained to be maximally-reward hacky, that took opportunities to reward hack within episodes, but never to increase the reward for other RL episodes.

I don't think Anthropic is taking these concerns sufficiently seriously. Which is extremely bad, because they're taking these concerns way more seriously than anyone else.

This story seems like minor good news, though they present it as bad news. The best explanation for this behavior is probably alignment to the spec, even if exhibiting this behavior in this particular scenario was net-negative.

Also, I think in the policy section, Anthropic should have advocated for an international AI slowdown or pause.

[feather-bone]

Charging forward.

[ID: an illustration of a bison running to the right on a grassy field. There are blue mountains in the background, and a cloudy sky. End.]

[mallowmaenad]

dark timeline: manosphere emu egg trend leads to new mainstream egg laying fetish as tiktok overwatchface model girls start roleplaying as having laid the egg the men are about to eat to encourage them to get their nutrients leading to an upward trend in yoshi kirigumis as niche sexual paraphernalia and a new character added to whatever hero shooter or moba is most desperate for new money at the time thats like sexy velociraptor themed with wide birthing hips and the manosphere gets over the emu egg trend when people start making gratuitous high fidelity SFM animations of that character laying eggs as it has now been deemed Gooner Shit to eat emu eggs causing emu farmers dealing with a sudden financial fallout and having to lean into that side of the appeal resulting in a new disgusting egg flavored gamer supps flavor featuring a leggy emu gijinka girl and people arguing online about whose waifu would have the better tasting eggs. Someone on twitter/bluesky/4chan makes a new "dinomama-chan" character who follows in a similar niche to the aforementioned raptor girl but with conventionally attractive young looking anime milf aesthetic who has a deadbeat NEET son who plays video games all day that she makes dino nuggies for as well as scrambled eggs made from her own eggs with all the expected sexual tension. The telephone game leads to trans girls getting better about eating breakfast and lunch by making themselves egg meals and imagining it came out of dinomama's cloaca making the original creator fabulously rich from plushie sales (not from makeship but the other one that lets you sell the plushies with sexualized designs I forgot the name) Hardly any of this is able to let emu farmers financially recover from the loss of sales due to the beginning trend raising the prices to being even more prohibitively expensive causing a politician whose really behind on the internet and culture who still thinks emu eggs give you big dick balls man milk energy lobbying to subsidize the emu farmers and going on hundreds of podcasts resulting in one of them eventually showing him AI generated oviposition porn on their phone making him think lizard people are real and hiding in our midst but like in a Super Mario Bros (1993) way causing another telephone game on bluesky and tumblr that jacking it to dinomama or owning dinomama plushies makes you antisemitic and the premier oviposition and lizardgirl fetish blog on tumblr gets nuked by staff after being mass-reported resulting in edgy transfeminist kinksters adopting an aesthetic around Riptor's cheerleader costume from the 2013 reboot of Killer Instinct in retaliation

[voxsymphonia]

huh

[aksemmi]

you wake up in a cold sweat. you have been reading about the history of ALF raids on fur farms. there is an emu farm a few hours from you. you know what you need to do. millions of hens cry out to you in your waking half dream. the egg industry cannot be allowed to expand it's profits