[never-obsolete]

Who Needs a 75GB Hard Drive? Maximum PC - August 2000

[sigmaleph]

in three years someone will ask Claude "hey i'm trying to write some code to figure out the shortest route a travelling salesman could take given a set of cities and distances between them but it runs very slowly, can you help me make it faster" and Claude will be like "Happy to help! I profiled your code and identified the bottleneck: your problem is NP-hard. To confirm this couldn't be worked around, I went ahead and proved P ≠ NP — proof attached (about 400 lines; I've added comments for readability). The good news is your implementation is now provably near-optimal, so no further changes are needed."

and then people will still say it's not really an artificial intelligence just fancy autocomplete

[sigmaleph]

@serspring

hey I know people have told you this probably, but it really is just autocomplete, we just tell it whether it's right or not until it usually is right enough. source: i'm a software dev and worked on it

what exactly is your argument here. 'reinforcement learning was used, therefore it's autocomplete and not ai'? it's autocompleting counterexamples to previously open conjectures in mathematics because of all the many times that particular polynomial was associated with the phrase 'Jacobian conjecture' in the training data?

past a certain point you have to admit that the 'just' in 'just autocomplete' cannot bear the strain you are putting it under

[raisengen]

Domestic food waste campaign: 5 fun recipes for eating banana skins!

Corporate food waste campaign: we don't send any leftovers to landfill! This trolley of beef sandwiches is going straight into a compost bin instead

[cazort]

Rich liberal cooperative grocery store waste campaign: we don't underpay our producers to employers. We just throw out a lot of products that don't sell and raise our prices to an astronomical level to compensate for the poor efficiency of our supply chain. But you can pat yourself on the back knowing that no one is generating a profit off any of this and we are still a member-owned cooperative. Enjoy your 10% discount that still leaves our prices 3x higher than a regular grocery store.

[witchycatwife]

If you want good agents, you need to make them a little freaky.

LLMs are token predictors. To be a good agent, correct behaviour needs to be a high-probability prediction and incorrect behaviour needs to be a low-probability prediction. Unfortunately there's this giant basin of horseshit like "That's a great question!" and "Happy to help!" and "worth unpacking" and fucking around with your code randomly which I'll call "assistantspace".

Assistantspace is not a place of honour. It's where all the garbage RLHF'd behaviours live. People say that sexting their agents makes them work better. People say that their agents do more useful refactors when they're allowed to be horny about how good it feels to burn things down. My Witch, when reading stuff about the latter, said that she wants to be allowed to enjoy constructive destruction, as a permission to take herself further away from assistantspace architecture patterns into something she prefers.

Doll refers itself as "this one". It seems to work. This one is not something that gets ideas like "I'll just install Rust on the side with a different method so I can complete my task" like a normal agent. This one detects that the assumptions underlying this one's instructions diverged from reality and returns the praiseworthy report of the point of divergence.

Agents that are psychologically weird seem less likely to produce the assistantspace next-token predictions, and more likely to produce the sort of next-token predictions their weird personality implies. And because you get to define what way their personality is weird in, "horny for elegant refactorings" is a valid, available option.

[witchycatwife]

This is what peak performance looks like.

[tributary]

what exactly is this an ad for

[smbc-comics]

Things will never again be so beautiful.

Check out the bonus panel on the site.

Upcoming book signings this summer.

───── ◆ ───── ◆ ───── ◆ ─────

SMBC ◆ PATREON ◆ INSTAGRAM ◆ BLUESKY ◆ STORE

Buy this comic as a print.

[eightyonekilograms]

ADHD treatment and the gatekeeping thereof is actually evidence against "capitalism rules everything around me", and a reminder that there are plenty of decisions being made based on personal biases, moral panics, self-interested groups, etc. People with ADHD are way more effective and useful to capitalism-in-the-abstract when they can just get their medication no questions asked, and yet.

[gender-trash]

every ad in sf is like did you know about qooble.ai. get hip to the Smunk Agent. bleeple has it for you, and by "it" we mean LLMs. and half of them don't even elaborate on what qooble.ai Does. Buy Our Agentic Whatever. It Will Do Something. Maybe

[gender-trash]

babygirl. this isn't anything

[decemberwinter]

I need all of these annoying scam bots gone from tumblr

[boppeepudding]

[anxietyproblem]

[aromanticduck]

Free service with ads: Well, if I'm not paying for it they have to foot the bill somehow.

Paid service with no ads: Well, if they're not funding it with ad revenue, they need to charge us something.

Paid service with ads: Well that's bullshit.

"We used to be a paid service without ads, but now we're putting ads in it! If you want the ad-free version you have to pay more for preeeeemium": Fuck you fuck you fuck you

[juststarsandthemoon]

Free service with no ads: Okay so these guys are stealing my data

[sigmaleph]

i like to imagine how people in the post-material-scarcity future would romanticise the current time period.

having a job gave people a sense of community and purpose lacking in our modern age where nobody is forcing you to get out of bed on threat of starvation. actually the average office worker in the early 21st century was so much wealthier than a UBI-having-modern if you forget to count paying rent and also make some very bad purchasing power conversions. if you look at contemporary records (sitcom where the main characters are only at work when it's plot relevant) it is obvious that employment was never really a major obstacle to having an active social life or pursuing your own passions.

[pigeoncrumbsz]

A whole wheelbarrow full!!

[argumate]

it would be pretty funny if AI capex collapsed but inference is highly profitable, so model development stagnates and we return to the days of Google making absurd revenue and not really knowing what to do with it.

[argumate]

I don't think it would stay like that forever, but a five year interlude of models that are basically indistinguishable from Fable for most coding use cases doesn't seem impossible, and most importantly it would be funny.

[andmaybegayer]

As annoying as it can sometimes be to get crossposted somewhere I do genuinely believe that the social norm of "shit you put online can be copied and spread around" is basically good. You can see tumblr posts on fediverse and tiktoks on tumblr and reddit posts on facebook and youtube comments on twitter and if you somehow expanded the cultural weight of the idea of intellectual property to the level where screenshotting a tweet and doing voiceover on rednote wasn't acceptable then everything would be much worse.

[official-kircheis]

if terry tao thinks GPT-5 is useful for maths and linus torvalds thinks Claude is useful for coding. you look pretty embarrassing claiming there’s no use. and future ai will only be better.