shambibble @shambibble.com · 6hat this rate, i'm gonna end up voting for the dem primary candidate who sounds the most like ed zitron, because "up against the wall for technically incorrect reasons" will be preferable to returning to any kind of business as usual with these oligarchs 050
shambibble @shambibble.com · 7hpangram is the only AI detector that has earned "trust me bro" status, and it does not appear in the article. applying the eye test to graphs of unicode points w/o a control can only ever be suggestive. what is also suggestive: the friendly interaction in benjamin ryan's inevitable first post 040
shambibble @shambibble.com · 30/09/2026crockett more or less confirming that this post-primary tweet will be her sole gesture of support and she will not utter a single positive word about talarico (this has been obvious for so long that frankly continued silence would've been preferable to whatever the fuck this is) 100
shambibble @shambibble.com · 30/09/2026well i can't say that happened to me either it was more like a whole lotta nuthin' 030
shambibble @shambibble.com · 29/09/2026excited to spend several thousand dollars on this and find out i'm a triple-nonresponder 180
shambibble @shambibble.com · 29/09/2026i don't think this holds universally, but there is a certain type of "legislating" that amounts purely to broad, lightly qualified delegations of power, and such bills i think would be a mistake to pass *right now*, in conjunction with the current SCOTUS, instead of 2029, after court reform 020
shambibble @shambibble.com · 29/09/2026if someone does i've got some real downers locked and loaded bsky.app/profile/sham... 110
shambibble @shambibble.com · 29/09/2026you're not controlling for the "is this an issue where either kavanaugh or barrett will feel strongly enough about it not to stay the inevitable lower court injunction against the inevitable overread" factor and given the NIH cases i don't really see the letter of this law as worth a pot to piss in 000
shambibble @shambibble.com · 29/09/2026i pulled the ripcord at 3d, which might be the single worst deployment of apophasis i have seen in my life 010
shambibble @shambibble.com · 29/09/2026"attempting lazy misandry and doing transphobia instead" is a failure mode i see from more and more big woke 1 accounts lately and i feel like there should be more learning happening 000
Reposted by shambibbleKylie Eleison @gonebabygone.bsky.social · 28/09/2026This is playing devil's advocate in the sense that it is the literal perspective of the devil during the temptation in the desert 120839
shambibble @shambibble.com · 29/09/2026yes, i am not generally going to believe "some random right-wing flack" on a law's cost-benefit regardless of any personal testimony by "some other random guy i've never interacted with." you are welcome to further replies telling me how awful my epistemics are and how great a lay he is 000
shambibble @shambibble.com · 29/09/2026PhD from george mason and an employment history of three right-wing think tanks? yeah, very happy assuming this is low-perplexity output that is far more AI-replaceable than a licensed cosmetologist. mouth sounds mouth sounds low taxes mouth sounds 100
shambibble @shambibble.com · 29/09/2026i don't know what the fuck makes you think i'm interested in what a Senior AEI Fellow thinks 100
shambibble @shambibble.com · 29/09/2026thank god this idiot isn't editing my city's newspaper or something 010
shambibble @shambibble.com · 29/09/2026yeah, sure man, the most anti-corruption president ever, who rolled back obama's rules on scam colleges in 2017, and who ran a scam school himself for several years, is definitely going for those guys and not "the liberal arts as a concept." 240
shambibble @shambibble.com · 28/09/2026i should note that much of the time this is a feint. scott alexander is not a "high-decoupler" on race; he views flippant racism in the sailer/hanania style as such a positive epistemic signal that a schizophrenic nut permabanned from his own discord for slurs is liable to turn up on his links page 41617
shambibble @shambibble.com · 28/09/2026my job (a medical school) has a ZDR provider but its perpetually several versions behind, given the pace of model advancements i could easily see them using personal accounts for similar reasons 020
Reposted by shambibbleJorvorskie Layne's @jorvorskielaynes.bsky.social · 27/09/2026Never in doubt 121
Reposted by shambibbleJ. C. Cantwell 🌻 @segfaultvicta.aludel.xyz · 27/09/2026as someone who has been really critical of a lot of the initial breathless "teehee, weird sex" reportage, i really appreciate this thread for nailing the actual fucked up immoral thing that is happening; you CANNOT consent if your employment or job success are on the line 8860199
shambibble @shambibble.com · 27/09/2026honorable mentions to utopia and full tilt poker as they are edge cases i am not sure qualify in the same sense as the others here 120
shambibble @shambibble.com · 27/09/2026i do not think the "rogue agent" vocabulary is particularly exculpatory to normies, if anything it makes the companies look more obviously incompetent. there is just no magical incantation that will suddenly make the executive enforce existing hacking law against a company that gave him $25 million 000
shambibble @shambibble.com · 27/09/2026i don't really think highly of bernie's Yud Act bill (which would have the immediate effect of conferring vastly more power to trump), but this paragraph doesn't hang together much at all. if anything x-riskers are overshooting the "take AI seriously" target, certainly compared to the word police 120
shambibble @shambibble.com · 25/09/2026no they haven't figured anything out yet, they just spotted an enzymatic sequence and ran off to breathlessly extrapolate it to "NEW CRISPR" 000
shambibble @shambibble.com · 25/09/2026i want to *steer* in the latent space, not roll for a code to a random spot (and their web-scroll style picker thing just never worked as well as the v5 tuner) 101
shambibble @shambibble.com · 24/09/2026my interest in MJ peaked a while back with v5's style tuner, fell off when they switched to gacha codes, and was buried when they killed negative prompting 100
shambibble @shambibble.com · 24/09/2026mfw anthropic hypes up another in silico bioinformatics result that would get a politely encouraging nod from your average PIstatic.klipy.comCam Newton Super CamALT: Cam Newton Super Cam 121
shambibble @shambibble.com · 24/09/2026i was specifically trying not to subdivide every letter (for realistically random breaks) but this is probably me underspecifying what was in my head here as well as there 010
shambibble @shambibble.com · 24/09/2026i eventually ended up doing a procedural cheat where gemini made a rule-based HTML mockup which i fed as an image reference (pretty much the only backup weapon in your arsenal these days) with instructions to make more organic, but i'm open to this being a promptable skill issue (for me or gemini) 010
shambibble @shambibble.com · 24/09/2026curious as to the best ransom note you can get of this length. i don't use GPT but was handing gemini two grafs of instruction and one graf warning against common foibles, and was still getting this garbage, which has superficial collage trappings but doesn't hold up to scrutiny 260
shambibble @shambibble.com · 24/09/2026that's the distribution you are noticing; it's difficult to get the GPT/Gemini models especially to be *imprecise* in an artistic way, they are heavily trained in an infographic/digital clip art pocket, and you have very limited tools to steer them to novel styles 070
shambibble @shambibble.com · 24/09/2026most AI image generators have buffed out any complexity and made it difficult to go out of distribution because they want their users to get "good" output from one subliterate prompt, which curtails the latent space and incidentally afflicts them with ben garrison-tier prolixity 3647
shambibble @shambibble.com · 24/09/2026mark my words "claude-as-judge" being the meta for approximately two years will haunt us for twenty 110
shambibble @shambibble.com · 24/09/2026i fully expect the coding models hiring the most cracked coders to get really good at coding, now they badly need to transition from alpo to mcdonalds 100
shambibble @shambibble.com · 24/09/2026rob sheridan's meat gala alerted me to the power of negative prompting in midjourney v2 and i haven't been happy without it since 000
shambibble @shambibble.com · 24/09/2026like bacterial resistance, such methods prevent training for one generation of AI models and then make the dataset richer for all future models! 000
shambibble @shambibble.com · 24/09/2026an incredibly :grimberk: unintended effect of oversold anti-AI art noisers is that, having persisted long past the models they had non-placebo effects on, you can by now straight-up improve the quality of some more modern AI outputs by negative prompting "adversarial noise" 110
shambibble @shambibble.com · 23/09/2026should be pretty easy to write a national security pretext for a law that says Congresspeople are only allowed to post with their own fingers 010
shambibble @shambibble.com · 23/09/2026I am struggling to think of any technology where the most interesting question it posed was "what to call it" 010
shambibble @shambibble.com · 22/09/2026i actually have no idea how similar the tics are in other languages, is there a lot of "ce n'est pas X, c'est Y" 030
Reposted by shambibbleJohn Timmer @jtimmer.bsky.social · 22/09/2026Ok, this "there are two separate brains" thing appears to be making the rounds, and I want to be absolutely clear: this is a press office overhyping some perfectly reasonable science that won't come as a surprise to most developmental biologists. 32754238
shambibble @shambibble.com · 22/09/2026confident in claude's ability to properly law, i approach the bench. "this is a verifiable reward just like math," i think to myself in a confident business professor voice as the judge slams their mallet directly into my skull 000
Reposted by shambibblehikikomorphism @hikikomorphism.bsky.social · 21/09/2026Nation-state and lone-actor misuse of LLMs really are serious threats, but that applies to the US state and to US billionaire lone actors too - if we're to seriously think about this problem, we can't pretend Elon Musk and Dept of War CTO AI lethalitymaxxing posts aren't in scope as concerns. 518321