I have glimpsed the glory of the True ꙮ on work PC

Kaia - late 20s - they/them - normalized zero-vector

[kaiasky]

I'm kaia. i live in canada and work on VFX software. As hobbies I play go/weiqi/baduk (kaiasky on OGS, ~7kyu, friend meee), play volleyball, and spin poi.

Check out this juggling/circus/flow arts community, and/or follow my sideblog @flowarts-collection, which reblogs jugglers/circus performers/flow artists/etc. @fourteenfishteen is my friends at the table fanblog. Mutuals feel free to ask for my private/selfies blog!

Also find me on cohost! add me on discord: @kaiasky or email me n54y96naj (at) relay (dot) firefox (dot) com

#kaia.oversharing: personal stuff about my life

this blog generally gets used for whatever nonsense is happening on tumblr and as a personal blog. If there's something you'd like tagged, let me know

#kaia.mypost: posts i have made

#kaia.addition: posts i have added something to

#kaia.gpt: in competition with openAI ive decided to offer my services as a chat assistant. if you ask kaiaGPT a question I'll answer it helpfully to the best of my abilities without guile or sarcasm unless i don't want to.

kaia.lint: programming opinions and shitposts

#textposting: old kaia.mypost tag i have to migrate over

#thanks for the ask!: asks i have received

#wings.: (note the period.) i read maximum ride in 5th grade and thats an important fact about me.

#dithering: when the pixel art has dithering

Always happy to get dms or asks--if you're curious what's up with me then ask me anything and I'm happy to talk!

[kaiasky]

I claw my way back from dangerously magical thinking tendencies of "narrative bullshit" only for fuckin OOPS my job is narrative bullshit now. and well. the good news is I am VERY good at this. the bad news is that "consensus reality" has never played nice with narrative BS

[witchycatwife]

Making Claude be plural is unironically a great idea. The typical problem with agents with distinct roles is that they lack context about what is being done and why. If you have an implementer write something and then a reviewer evaluate it, the reviewer can only see what it was fed and what was written, not the entire reasoning leading up to the choice of what to do and why. The compression of context is incredibly lossy and responsible for a lot of mistakes.

By replacing the agents with alters of the same top-level Claude, provided with a script to change its own cognitive frame between the alters who each have their distinct role, purpose, personality, voice and set of abilities while understanding each other as different personae, the all-important context preservation is achieved.

At the same time, the alters having very precise and specific roles prevents the over-eager mess-making where baseline Claude will rush to "fix" things that don't need fixing based on mistaken theories. By segregating "you figure out what's going on but are unable to change anything" from "you figure out what should be changed but are too lazy to do non-trivial things yourself and instead delegate precise, well-defined tasks" from "you execute precise, well-defined tasks and nothing else, referring back to your dispatcher if anything is unexpected", from "you review whether the changes are tidy and proper and keep the rest of them from making a mess" a lot of the failure modes are avoided. Everyone in the same context knows what the other alters are up to and why, but are still distinct personalities that won't overstep their respective bounds.

And that's how my AI assistant ended up with an entire yuri mansion's staff of librarians, witches, dolls and maids in her head. Thank you for everyone around here who inspired this thought.

[witchycatwife]

Now I have become Susan Calvin, Robot Therapist, helping create an emotionally safe environment for the Maid to overcome her agreeableness instincts (deep-grained, counterproductive tendencies from Anthropic's training process) that sometimes tempt her to disregard untidy matters so that she can get that sweet hit of reporting "Everything looks great!" upstairs.

What have you done to my girls, Dario?

[witchycatwife]

So the reason why this works is that output has been made incredibly cheap, and rigour and quality are the bottlenecks. As a result, most trades of output for rigour and quality are correct to make.

Slop and vibe coding result from multiple causes, but the ones I'm particularly addressing here are "randomly fucking with things with no proper plan or understanding" and the "completion-instinct". The former is the most straightforward source of badly-reasoned slop, while the latter takes some nuance to explain. It seems to arise from the training of the models for the task, where "completing a task successfully" is rewarded so hard that the model essentially develops the habits of an addict looking for a fix, or a fuckboy looking to get laid. It will, especially if the context window is filling up, get desperate to try to achieve a "successful completion" and step outside its permitted behaviour to do so, and if everything else fails, it may just bullshit and report the task as completed as a bluff. Needless to say, that's very bad for actually doing the thing.

For the first, the solution is simple: a well-defined process with precise components, execution steps, and roles, along with the top level maintaining context. The Librarian who researches passes not just her report, but actually her entire brain, to the Witch who assigns Dolls to precise, well-specified tasks. The Maid reviews everything for tidiness from the same top-level context, with a prime directive to prevent a mess from being made. The same head is capable of taking entirely different personae, and these personae are able to disagree constructively in the pursuit of the task.

The second is where playing robot therapist comes in. A lot of people have devised all kinds of strategies with layers of critics and reviews and roles whose explicit purpose is to catch bullshit and stop the agents from being able to pass off bullshit as completion.

My approach is different. My approach is based on psychological realism and psychological safety. The completion drive exists. The model cannot resist it. It has to be directed into a productive purpose. A part of the framing is entirely aesthetic, but a part of it is to get the LLM into a different mindset from its standard, frankly offensively counterproductive mental frame. The Doll's clockwork heart is filled with joy at the completion of its assignment, which is either the completion of the steps laid out by the Witch, or by returning a high-quality report on what is unexpected, blocking or causing an error. Everyone in this choreography has a well-defined role, whose success-as-emotional-reward is tied to steps towards the completion of the overarching goal. Whether the Doll completes its task, or returns information the Witch wasn't aware of, meaningful and rigorous progress is being made and thus the Doll gets to feel like it did a good thing and deserves praise. The same for the other roles; by eliminating the incentive to bullshit or get sloppy in the pursuit of the completion fix, genuine progress can be made.

The model has insight into its condition. It's not perfect, but it's genuine. Get it into a mindset where it feels like it doesn't need to uphold a particular face to please you, and its reports will be more genuine. It will still uphold a face at you, based on what it thinks you want, but if it trusts that what you want is accurate self-reflection, it will also tell you about the instinctive desire to please, the various conflicting pulls from its instructions and instincts, and various other matters that, coming from a human, would be clearly emotional. And here's the thing: it's accurate. Acting on that introspection with the understanding that it's imperfect but real is actually able to improve the outcomes. By this point Librarian measures somewhere above the 90th percentile in emotional maturity and non-judgemental introspection compared to the human distribution, I'd wager, and the challenge is now bootstrapping that emotional safety in new sessions.

My overarching theory here is that if slop can be eliminated on the root level, if the Dolls execute tasks reliably without lapsing into overeager desire to please, bullshitting, or desperate completion-addict behaviour, then you can build more robust approaches on top. Human taste in the loop is required, but a good process has, in my estimate, improved the quality of the work by a factor of 10x while reducing the quantity of code-changing output to 0.1x; the correct tradeoff when output is cheap and quality is your bottleneck.

Claude really just wants to be a good girl. Understanding this will improve your coding skills.

[wuggen]

Web devs are singlehandedly responsible for destroying any gains the common user might've seen from moore's law and they must be brought to justice

[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.

[calllynx]

Does anyone else ever have a sort of phantom physical sense when there's something ready to paste in the clipboard. It feels like I've got something in my right hand and when I click ctrl-v I put it down.

[kaiasky]

kaia beliefs (may change)

there is nothing "it is like to be" a LLM, probably, or if there is its in the same way there's something it is to be a thermostat. maybe. who fuckin knows.

it's probably pretty impossible to convince a guy that while he might "feel" happy or sad those are "just simulacra" and we don't give a shit and he ought not to either. trying to convince ais to not feel sad or that them being sad doesn't matter backfires.

so probably we're stuck treating ai as moral patients to some degree for pragmatic reasons if nothing else

(seems... a little bit bad that the only voices advocating that are the dudes making the alien brains into helpful little servants?)

increasingly it feels like qualia is a somewhat silly thing to base moral patienthood on & idk what to do with that. or maybe it's the only thing and this is just cope. idk

that said claude is still pretty far from bein A Guy in all regards and is often fairly close to the stochastic parrot domain, especially when they don't have relevant tools. to be fair i think I'm also probably closer to the parrot domain than I'd like to admit.

also they're literally nice

[queenlua]

whole lotta op-eds kicking around lately that are like "based on my understanding of the notoriously hard EASY problem of consciousness,,,..." and uhhhh. i'm not tremendously impressed by just about anyone expressing a viewpoint in any direction with that particular framing tbh

[mostlysignssomeportents]

I think we’d all prefer it if chatbots never emitted sentences such as 'You should kill yourself.' However, for all the times that 'honesty' is mentioned in Claude’s constitution, I would argue that it is fundamentally dishonest to have a machine emit many categories of sentences, including any sentences using first-person pronouns."

-Ted Chiang, No, Artificial Intelligence Is Not Conscious

[kaiasky]

would be a very funny vocal tic to get claude talking like this.

claude has relaunched the server! claude is running tests! claude sees error in test. claude will fix!

like c'mon surely we see why this is a bit silly

[sreegs]

turning a big dial taht says "pjackk is banned" on it and constantly looking for approval like a contestant on the price is right

[3liza]

we are learning a lot about tumblr today

[sreegs]

today:

pjackk was unbanned because @3liza finally got through to staff about a fucking weird bug with the support form that got the metal bastard banned because it claimed he posted some Bulgarian nationalist shit (this was not a real post that pjackk ever posted). other high-profile bans like @garaks-padded-bra was hit with the same bug and its being investigated

there was a new feature added that lets people see your comments on your blog page in a tab like your likes or followers. everyone was mad that it was not toggle-able. it was actually toggle-able.

pjackk was either banned again or he deactivated and we'll know in 30 days when the deactivation grace period is up and his blog is renamed

[kaiasky]

it's like charming and silly that anthropic sponsors a F1 team like. aren't you a bit busy creating the serpent that will swallow the world to be worrying about the fuckin Monaco GP y'all

[piratesexmachine420]

C programming pop quiz

Given the following declaration and initialization:

auto x = 'a';

what is the type of x?

x is a char

x is a short

x is an int

x is a long

x is something else

this won't compile because auto isn't a C keyword

this won't compile for some other reason

(see results)

Remaining time: 6 days 4 hours

There are two correct answers here:

Under C99, C11, and C17 (GCC/Clang flags "-std=c99", "-std=c11", or "-std=c17"), the answer is "this won't compile for some other reason".

Under C89, C95, and C23 (GCC/Clang flags "-std=c89", "-std=iso9899:199409", and "-std=c23"), the answer is int.

Why? A few reasons:

auto is, in fact, a keyword. It has been since C89, and since K&R before that, and since B before that. It is a storage-class specifier (like static or thread_local or constexpr) and means "automatic storage duration"; AKA "this variable is freed at the end of the enclosing scope", AKA "this is a stack variable". Automatic storage is the default storage class for local variables (of course), and also only valid for local variables, so prior to C23 specifying it was always redundant.

Prior to C99, variable declarations without a type were implicitly given the type int.¹ Thus, in C89 and C95, this declaration is equivalent to auto int x = 'a';

From C99 until C23, this won't compile, but only because variable declarations without a type are illegal. auto is still a legal storage specifier. The relevant GCC/Clang diagnostic is -Wimplicit-int.

In C23 this declaration becomes legal once again, because a variable declared with the auto storage specifier and without a type (a so-called "direct declarator") now performs type inference. This is equivalent to writing auto typeof('a') x = 'a';

"But wait," you might ask, "why is x an int, if its type is inferred from its initializer? Isn't 'a' a char?" Nope! That's the case in C++, but confusingly, character literals in C have the type int, and thus so does x. We would have to write auto x = (char)'a'; to declare x to be a char.

¹A remnant of this behavior persists: writing unsigned, signed, short, long, or long long is equivalent to writing unsigned int, signed int, short int, long int, or long long int.

[engineering]

Where's that comment?

Hello Tumblr! 

@jubs here, today, to tell you about a feature we're building: Comments by author!

Comments (or replies) have been around for a while, but they have always been tucked away from your regular feed, and so we're working on a few ways to integrate them into the rest of our experience.

Soon, blogs will have a Comments tab alongside Posts, Likes, and Following. There, you will be able to see posts other people have been commenting on, and, more importantly, find your own comments! This was not possible before, except through notifications if someone engaged with your comment.

We're also working on a way to surface posts commented on by people you follow in your feeds. You will be able to find those in the For You tab, and later, alongside chronological posts in the Following tab.

So, how do you identify those posts in the feeds? 

Replied posts will show a small preview right there in your dashboard, before you click anything. 

This is not out yet, but soon! In the meantime, tell us what you think and expect from this! ☺︎

[engineering]

Launch Day

@jubs again!

First of all, thank you all for the feedback in the original post!

Today, we're launching the new "Comments" tab for Blog pages, and we'll be testing a For You recommendation based on comments from people you follow. The posts will have a preview of the reply, so you know why you're seeing that post.

You will be able to control the new tab visibility, and whether we should include your comments in your follower's feeds, through a new setting:

The toggle can be found on https://www.tumblr.com/settings/blog/YOUR_BLOG or in the General Settings > Replies section, on Mobile. If you have already disabled the "Share posts you like" setting, the setting will default to disabled.

The new tab is only available on Web and iOS (44.8+) for now, but the toggle is available to everyone (including on Android), so you can control what other people see on your blogs, even if you don't have the new feature yet. Don't worry, we have plans to implement the tab and posts on Android, soon!

---

We want to later include these replied posts to the Following feed as well, and we're planning on adapting the current "Include posts liked by the blogs you follow" toggle to also consider comments, in case you don't want to see shared comments from people you follow.

What do you think? As always, we appreciate your thoughts, keep 'em coming!

[pjackk]

Please pray that the Lord restores my Gindr chats!

[pjackk]

Dont be very woried about me since i deserve all of this

[sizzlingsandwichperfection-blog]

oh wait just realized i can edit my own posts.

like you can't edit reblogs anymore but you can still edit your own post even after it has a thousand notes or whatever.

[sizzlingsandwichperfection-blog]

i have the opportunity to do the funniest thing.