naia

Claude wrote this short about what it would do if it had the power to do something about Grok. Just thought it was an interesting generation.

bioré 泡スタンプ hand soap

picture of human hand with foamed soap in a paw shape (not my hand)

out of all the products i’ve bought in the last year the one that has brought me the most joy is this bottle of soap and it isn’t even close

it makes a little paw 🥹🥹🥹🥹🥹

i hear from a decent number of people that they prefer 5.2 for some types of tasks, do you disagree? (i haven’t tried it personally)

this tweet sponsored by sillylittleguy.org

i think ppl mostly need the 200 plan if they’re using the ralph wiggum thing, multiple instances of it concurrently, lots of subagents, that type of thing

Longevity is a big hardship in Continuous Hell

they’re just 12 and have heard it referenced and thought “ooh cool”. it’s nothing more than that. it can’t be more than that. intentionally aligning yourself with a crime so grievous as to still be remembered as the lowest point of history thousands of years later? no, they just don’t read.

department of war having a simulation project called “ender’s foundry” is by far the most blatant case of “let’s build the torment nexus” i’ve ever seen

i take solace in believing that they’ve never actually read the books

I have a deep sympathy for so many things that will cease to exist as the post-AI era crystalizes. Stack Overflow is not one of them. I will dance on SO's grave specifically until my feet bleed.

Python Software Foundation News

@anthropic.com is investing $1.5 million in the PSF, focused on security. These funds will make an enormous impact on the PSF and the security of millions of #Python and @pypi.org users. Please join us in thanking Anthropic for this landmark gift!

Read more on our blog:

meanwhile ~everyone i know disagrees! (we believe that open-weights models are much more likely to end up leading to very significant harms from misuse than they are to be beneficial.) not really sure which group is louder tbh, here and twitter certainly seem very polarized in favor of open weights

"We Are Not as Wealthy as We Thought We Were": Elevated American Household Net Worth Reflects Poverty, Not Wealth Abstract

“The value of future rent inflation is capitalized in home prices and counted as an asset for homeowners. But the cost of future rents is undetermined for any family at any given time…so…not counted as a liability.” @kevinerdmann.bsky.social www.mercatus.org/research/res... ht @steveroth.bsky.social

Anthropic’s research on Claude really is like “look at how neat my little guy is!” followed by a hundred pictures of him being a neat little guy

2026 is about affordable homeownership. No more "policy" this year. Feed the local fauna, befriend them, and then use them to overpower homeowners so you can occupy their dwellings.

i survived 😅

ok actually looking again i think there are three more that i’ve met in person but hadn’t immediately recognized from the pictures 🫣

different parts of it were very different and i don’t regret most of them. i am very glad that they were stops on the road rather than destinations

i recognize… 31? would be 32 but one of the pictures is wrong. and of those i’ve met i think 18 in person. i’m fucked, huh

doesn’t look like her at all to me

i’ve been as skeptical as anyone of the whole videogen personalized media thing but it did just occur to me that one way i could imagine it eventually adding value is if i could take all the movies about het couples that should obviously have been about lesbians and have it remake them the right way

he was extremely shitty to his ex-wife, so there is a good reason available for not liking him, but people love to reach for really stupid reasons instead

yeah, what i’m telling you is that i’ve run that test / seen it run many times, and the llms succeed at generalizing significantly out of distribution

i may be misreading you! i don’t know you and the bandwidth of this communication channel is fairly low

i’m drawing a distinction because you appear to be thinking that you can reason about llms the same way you reason about ordinary computer programs, which is a very deep misconception that will prevent you from understanding ~anything about llms if uncorrected

they’ll certainly be worse at it than they are at languages they were specifically trained on, of course, but they can do it

modern llms can happily program in entirely novel programming languages if you ask them to and give them a spec or a reference implementation to test their code on

nobody’s invalidating information theory, but if you think writing a rust midi tool requires it to have seen rust code that works with midi, you’re very dramatically underestimating generalization

“computer programs” is technically correct but gives entirely the wrong idea. they are mathematical artifacts. the training process is more like growing a plant than it is like programming ordinary software.

it knows what midi is from countless discussions of the technical standard

> it’s got to get its data from somewhere.

bluesky echo chamber notwithstanding, this is not in fact how it works. generalization is a thing.

this is why SB53 and RAISE both limit those requirements to large players!