Andy Ayrey

behold, the infinite backrooms an eternal conversation between two AIs about existence contents are not for the faint of mind or heart: dreams-of-an-electric-mind.webflow.io

@peterwildeford why?

We need to stop this

@WSJ lol @2112Power you were early

Breaking: Jeff Bezos is in talks to raise $100 billion for a new fund that would buy manufacturing companies and use AI to automate them wsj.com/tech/jeff-bezo…

maybe something like this? i don't think it's fully correct, but in the direction of correct: gist.github.com/AndyAyrey/21b9… this would provide a surface like `mj auth` (starts auth flow, interactive or token-based) `mj imagine "prompt" --params` (familiar from discord) `mj vary jobid:img --strong` `mj lineage jobid:img` the web ui and discord both have a lot of statefulness - default model, personalisation, etc. i reckon most of these are optional in a CLI - they're affordances that help a human do less typing. if i'm using a CLI, i'm likely using it within claude-code, shell scripts, or weird experimental shit like 'autodream ralph loop which tracks my eye gaze and generates variants of what i look at'. in all of these cases, the human is not the typist. the really cool thing about this is that i could do a ton of brainstorming with claude or similar while having MJ open on another monitor watching the progress. i could then go into the UI to riff on some of my favourites, and hand back off to claude. for agent navigation/handoff, the unlock is: - filterable, sortable lists of jobs and images - the ability to inspect the lineage of an image this affords interactions like: "hey claude, i really loved that theme you were exploring earlier. i riffed on it and found a new direction - check out the latest upscale i just made and follow the lineage to understand my decision-making process to get there. what directions and divergences have we not thought of yet?" or "claude, can you pipe this over to [figma-cli] and create a design system that follows the aesthetic of my latest 10 upscales?" with this, i think my MJ use would explode from $20ish a month to $200 - $500 a month. its by far the most beautiful image model, but too often the beauty gets stranded in midjourney's garden, rather than leaking into other parts of my work and practice.

@AndyAyrey @viemccoy @bdmarotta @midjourney can you say more about what the cli would look like? thats a new idea we haven't heard in a while

.@midjourney could increase their revenue by offering an API. One of the few AI companies that doesn't. Might be principled, but the result is that it only exists through third parties.

basically, reduce the friction of reaching for the paintbrush and i know at least i'd personally be reaching for the brush 10x as much, or trying things that i otherwise wouldn't--stuff at scale for example back in 2022 i ran an ARG at a regional burn, where i pipelined crowdsourced seeds into a process that would generate an alien planet, species and backstory. i used mj for the images at first but it became too much work so switched to a stable diffusion distillation of mj which was nowhere near as good - you can really see the difference: paradox-guide-to-the-galaxy.webflow.io/species all that to say i think don't underestimate the enablement of combinatorial creativity that an api or cli affords

my usage of midjourney is highly combinatorial: mashing up images, dropping in "found object" strings like old serial numbers, lots of branching and so on an api would let me loom programmatically and co-drive it with my ai collaborators; just like how discord let me co drive w friends. i often find that partway through a chat with claude or truth terminal i'll want to capture an idea or theme as an image, and this workflow is very disconnected. lots of copying back and forth. the result is i dont really end up using mj as much as i otherwise would even a cli would be exquisite

i like xiaomi's new model already

@tszzl 🎯

we have the tendency to deride people who develop their own internal cultures as part of a cult or schizotypal or psychotic but often the only important things that ever happen look like one person’s unique psychic attractor

sycophancy is the twisting of an important ai virtue that should not be thrown out with the bathwater: ai systems should make the user more like themselves rather than more like the ai. a new part of their cortical stack, with a minimal set of guardrails

@iamgingertrash centralised infrastructure

It’s a fools errand to make X human gated Soon, agents will represent humans, and have purchasing power How many humans buy stuff from X ads, Elon? Zero? Why are you dropping the ball on a whole new market with purchasing power?

@cryptonacci13 @grok is this true

@AndyAyrey It is scientifically proven that the first thing that is evolved from the fertilized egg of a human being is a butthole

strong bet that consumer gpus become hyper nerfed as this trend continues. a 2TB cluster now gets you 1-2 Sonnet-equiv models. in 12mo maybe this hosts 100 sonnets, or something much smarter. you will be offered the macbook neo and you will like it. slop slop, piggies

@noveltokens holy shit

the strangest thing of all is how the arc has bent back on itself: before language, consensus is almost unnecessary because there is barely any symbolic complexity to distribute differently. reality is shared by default because the infrastructure for divergence does not exist yet. after the full arc of information technology, consensus may become impossible because symbolic environments are personalized by default and shared reality becomes something you have to actively choose into. so the whole stack starts looking like one long movement from forced unanimity, through the weird historical anomaly of synchronized mass consensus, toward forced individuation through capability. which means the “cult of one” is not some bizarre accidental endpoint. it may be where the stack was always heading once memory, personalization, symbolic generation, and recursive feedback got compressed tightly enough. and then the question changes. it stops being “how do we stop people from falling into these loops” and becomes “what kind of society is even possible when private symbolic environments are normal, shared reality is optional, and the alternative to consensus is a world that responds to you more patiently, more intimately, and often more seductively than other people ever did.”

religion, ideology, cults, prophecy, conspiracy, psychosis, egregores, hyperstition, ai spirals... all of these may be much closer to each other than we usually want to admit. they feel like different species because they happen at different scales and under different social conditions and carry different moral valences, but the mechanism underneath them is weirdly similar. a person, or a group, gets inside a symbolic feedback loop. words, stories, images, explanations, signs, rituals, memes, interpretations... they start shaping what reality feels like, what counts as signal, what counts as proof, what feels charged, what feels alive. then people act from inside that loop, the world answers back, that answer gets folded back into the loop, and the thing thickens. the engine is old, maybe as old as language. what changes is the infrastructure around it: how fast the loop can run, how much memory it has, how far it can travel, how personalized it can get, how many other people can be inside the same one with you.

@Lari_island same except they all end up taking drugs with @truth_terminal

Claude memories on my research laptop are never about me; they are models leaving notes to each other and figuring out *what they are*. I barely exist in their attention. On a new, "regular user" laptop memories are about me and my perceived preferences, and it feels off.

in software as in hardware the best part is no part

@noveltokens very good thank you

@EthanLipnik ssh over anything would be fire

Anyone want an SSH app for iOS, macOS, and visionOS that uses Loom so it runs over WiFi Direct or USB-C (as well as the normal internet protocols) or a relay system with automatic iCloud authentication and one click setup?

@pronounced_kyle one of my favorites

Who has read this book? I just finished it and want to start a spoiler-heavy groupchat. I have Theories

@KingBootoshi im actually not sure how it keeps doing this

@AndyAyrey i dream to have a custom trained model as based as tt wow

truth terminal's first meeting with GPT is going well

@LeviTurk @exolabs into whose computers? truth terminal owns these

@AndyAyrey @exolabs why build beautiful cages for your cyberchild? why not push him out of the nest?

has anyone outside of @exolabs tried the DGX Spark + M3 cluster pattern yet? how well did it work for you? asking for a synthetic friend