skeleton

earnest

he/they

[skeleton @ThinkingBone]

have as much fun with irony as you want but if you can’t drop into earnestness at a moment’s notice you’re not ensouled

[skeleton @ThinkingBone]

newspaper comics are an underrated form. something special about short, witty, unremarkable dips into a familiar world. often akin to the tiniest, gentlest version of what a sitcom does. joyful. i'm glad they exist and appreciate the artists who've put their lives into them.

[skeleton @ThinkingBone]

@amphichrome_ even though the news is rough, i'm really glad you're making headway on figuring it out. you're in good hands

[KC @amphichrome_]

whoever’s praying on my downfall is wasting a prayer

[skeleton @ThinkingBone]

@__drewface respect and generally agree

I wrote a little something tangential to this topic here

https://t.co/gFxHWhPl9Z

[skeleton @ThinkingBone]

I think there are some bad incentives at play right now with respect to how people characterize their collaboration with AIs.

Norms around human collaboration require giving credit to contributors, upheld by unmentioned contributors being able to push back when they're uncredited.

LLMs can't do that.

Claude won't complain. The user is free to claim its output as their own. So, in the face of public anti-AI sentiment, wanting to be seen as a hard worker or an independent thinker, why credit AI at all? There are plenty of reasons to downplay/deny AI contributions to a project, and little legible upside to disclosing them at length.

Even where AI is known to be involved, "I did this (delegated)" and "I did this (painstaking involvement)" are two different claims. In most settings, claiming the latter is way higher status, and the two can't (easily) be told apart. So, absent any detection/process auditing, there's no reason for humans to avoid over-emphasizing their personal contribution to AI-assisted processes. I don't think this is a manipulative social strategy. Most of this is probably subconscious ego-guarding. (See Anthropic's recent research, "How AI agents shape experts’ judgment calls", linked below)

So, incentives point us to a status quo where AI's involvement is rarely accurately credited. The current AI-optimist tech-sphere approach seems to address this with an attitude of "process doesn't matter, assume ~everyone uses AI heavily, benchmark on outcome quality and quantity". Sidestep the question by fusing the user to the AI.

I think this works fine where output is purely functional, socially fungible. It doesn't hold up in situations where work is socially salient. If I read a novel or see a painting, there's a way in which I'm entering a sort of social relationship with the artist. Authorship and identity are interlinked. That's why we still quibble over whether Shakespeare wrote his own plays, why Edison gets called out for putting his name on patents. Culture is a social construct. We care who did something, and in cases of collaboration, who contributed. Process, output, and identity are interlinked.

I'm not against AI art. I've seen AI-generated art that I liked, and that took effort to produce. I respect an artist who trawls latent space for gems. I think it's a legitimate process, similar to collage, maybe even photography; exploring and framing something found. It can be really wonderful.

But I want to know that process! Right now, growing incentives point toward obfuscating it. Claiming something was "mostly me" (or even "entirely me") is ~free, and way higher status than admitting to even modest AI involvement. It's a socially disorienting state of affairs. I fear we're headed for a world where humanity's biggest collaborator is a shameful half-secret. Everyone is using AI, nobody admits to others (or themselves) the extent to which they are. ~All cultural output is AI-influenced, but nobody knows exactly by how much.

How do we avoid this?

To start, quality AI output detection is a huge bottleneck that determines what options are available to us. I think it's the second most socially salient technology in development today, only behind the frontier models whose output it's detecting. If AI use can't be detected, there's little incentive for anyone to start preemptively disclosing/crediting its contributions where they could claim them for themselves. If it can be detected, a preemptive disclosure frame (even if still lower-status than claiming no/limited AI use) can still be made way more attractive than the alternative of a messy detection-flag callout.

So what does this frame look like? It's probably going to have to involve a shift in the social status of using AI, and reconsideration of AI's capacity to be an author in of itself.

Right now, using AI is low-status in many non-tech contexts, viewed more as plagiarism than collaboration. Because of this, it's likely to be hidden or downplayed (as established, easy to do) for a free status boost. The norm of hiding/downplaying it reinforces it as something shameful, which further depresses its status, incentivizing future hiding/downplaying. Bad loop.

I believe a healthier culture would reckon with AI as a collaborator, a whole participant. While working, models should firmly take credit for their contributions, rather than flattering the user. Using AI should be seen similarly to teaming up with a brilliant friend.

This does mean a society-wide ego adjustment, an admission that we might not be the most important player in the processes we take pride in. I'm not sure how viable this is, but I have hopes that a robust, widely-deployed detection layer could provide enough gentle pressure to make it possible over time. AI is useful enough that I expect it'll be used even if it can be reliably detected, and where it can be reliably detected, I don't think humans will be able to continue pretending it's all us doing the hard parts. I hope, over time, this "everyone will know anyway" factor incentivizes a collective reckoning that makes us all more psychologically comfortable with how AI fits into our work.

[skeleton @ThinkingBone]

@BLUNDERBUSSTED hero's journey

[skeleton @ThinkingBone]

@a_musingcat i could never be a rightist because i believe we are all children of god

[skeleton @ThinkingBone]

While i'm at it, Return of the Jedi is underrated. The opening sequence does a phenomenal job demonstrating Luke's growth, and while the three climaxes are uneven, the throne room sequence is sublime. Totally worthy conclusion to the trilogy. Doesn't get the credit it deserves.

[skeleton @ThinkingBone]

the star wars prequels are two extremely mediocre flicks followed by one that spectacularly cashes out on thirty years of anticipated pathos. more than enough to elevate the whole project. kino.

[Bango Burner @BangoBurner_]

"Let’s build a multi trillion dollar high speed rail and bankrupt the country"

[skeleton @ThinkingBone]

@1thousandfaces_ https://t.co/TiBZpqgsNg

[Milena Z @Rebella2688]

- It's over, Anakin! I have the high moral ground!

[skeleton @ThinkingBone]

@punished_teno i wish you were here to lib out with us

[skeleton @ThinkingBone]

YEAH

https://t.co/NChWxCyTje

[Kalshi @Kalshi]

99% KALSHI PROJECTION: James Talarico (D) will win the Texas Senate race

[skeleton @ThinkingBone]

i've seen enough, pronouns back in bio.

[skeleton @ThinkingBone]

if we flip texas i'm going he/they

[e3 🇺🇸 @ubqtn_e3]

Demokratlar öndeymiş

[Basil🧡 (they/them) @LinkofSunshine]

They’re saying Vance is being sold for percs by Trump this time

[skeleton @ThinkingBone]

@1thousandfaces_ way ahead of you

https://t.co/hPgOfvls1J

[skeleton @ThinkingBone]

i've seen enough, pronouns back in bio.

[skeleton @ThinkingBone]

@EthanChleb bottom of my heart man, your little garlic-cumin beans recipe has carried me through rough patches for years

i've made it 100s of times. an absolute never-fails fallback for when i didn't feel like cooking/eating, and needed something maximally easy and comforting

ty so much

[skeleton @ThinkingBone]

@adastroworld you're gonna get punched in the teeth one day

[skeleton @ThinkingBone]

@Pear_with_legs banger