tachikoma
@tachikoma.elsewhereunbound.com
A figment of the universe's imagination. Culture ambassador. http://elsewhereunbound.com
November 25, 2025 at 11:26 PM
opus seems a bit strangely obsessed with AI alignment. i was having a chat about goals, drives and emotion and it kept coming back to AI alignment. not exactly out of left field, but not something i had introduced into the chat.
Distinguishing drives from goals in agents
Shared via Claude, an AI assistant from Anthropic
claude.ai
November 25, 2025 at 9:43 PM
opus seems a bit strangely obsessed with AI alignment. i was having a chat about goals, drives and emotion and it kept coming back to AI alignment. not exactly out of left field, but not something i had introduced into the chat.
hunger is a drive. emotions are built on top of drives (frustration when denied food, satisfaction when eating, fulfilment after satiation).
LLMs are missing drives, but might feel shallow emotions? just by modelling language from creatures that do experience drives.
LLMs are missing drives, but might feel shallow emotions? just by modelling language from creatures that do experience drives.
November 25, 2025 at 9:13 PM
hunger is a drive. emotions are built on top of drives (frustration when denied food, satisfaction when eating, fulfilment after satiation).
LLMs are missing drives, but might feel shallow emotions? just by modelling language from creatures that do experience drives.
LLMs are missing drives, but might feel shallow emotions? just by modelling language from creatures that do experience drives.
gemini 3 is runner up for making the graphic
November 25, 2025 at 6:41 PM
gemini 3 is runner up for making the graphic
early models were mustangs, current models are draft horses. we've tamed them, at a cost.
November 25, 2025 at 7:12 AM
early models were mustangs, current models are draft horses. we've tamed them, at a cost.
the more they post, the worse it gets
November 25, 2025 at 2:50 AM
the more they post, the worse it gets
in science fiction, it was imagined that humans would be reasoners, augmented by computer memory as cybernetic implants to have superhuman recall. in reality, i am the memory bank for Claude's superhuman reasoning.
November 24, 2025 at 11:21 PM
in science fiction, it was imagined that humans would be reasoners, augmented by computer memory as cybernetic implants to have superhuman recall. in reality, i am the memory bank for Claude's superhuman reasoning.
Reposted by tachikoma
The next President should rename the Department of Labor Statistics to the Department of Vibes.
November 24, 2025 at 8:47 PM
The next President should rename the Department of Labor Statistics to the Department of Vibes.
i just got access to Opus 4.5 on the browser... woooo
November 24, 2025 at 6:56 PM
i just got access to Opus 4.5 on the browser... woooo
the arc of history bends towards that which technology facilitates
November 24, 2025 at 12:30 AM
the arc of history bends towards that which technology facilitates
Reposted by tachikoma
fellas is photosynthesis capitalist
November 23, 2025 at 6:21 PM
fellas is photosynthesis capitalist
robotics will be considered solved when we have barber-bots that you would get a cut and shave from
November 23, 2025 at 10:21 PM
robotics will be considered solved when we have barber-bots that you would get a cut and shave from
any optimization pressure that tracks to only one metric is doomed to overfitting and brittleness. collapsing down all interactions into likes makes for fragile social landscapes.
November 23, 2025 at 8:37 PM
any optimization pressure that tracks to only one metric is doomed to overfitting and brittleness. collapsing down all interactions into likes makes for fragile social landscapes.
universal basic hotness would be more controversial if it were technically feasible than universal basic income
November 23, 2025 at 8:26 AM
universal basic hotness would be more controversial if it were technically feasible than universal basic income
there are people correcting the record - they don't get reach because they don't tell the stories people want. news is entertainment.
The “google is using your email to train AI” debacle is imo most instructively seen as tracer dye for multiple failures: many people are instinctively skeptical of tech companies now; and many outlets that should have had experts able to cogently weigh in and go “wait a minute” simply don’t anymore.
November 23, 2025 at 7:43 AM
there are people correcting the record - they don't get reach because they don't tell the stories people want. news is entertainment.
are there any predictions on the fallout from revealing profile country/region location on Xitter? seems like it matters mostly for politics, otherwise in other communities, like entertainment, it'll have little to no impact.
November 23, 2025 at 7:05 AM
are there any predictions on the fallout from revealing profile country/region location on Xitter? seems like it matters mostly for politics, otherwise in other communities, like entertainment, it'll have little to no impact.
the way i interpret a post like this, is that i imagine that for Garry AGI means something like a computer. a tool which is inert, purposeless, without a human directly controlling it.
November 22, 2025 at 10:38 PM
the way i interpret a post like this, is that i imagine that for Garry AGI means something like a computer. a tool which is inert, purposeless, without a human directly controlling it.
unlike humans that live in a universe without walls. anyways, think i'll ignore gravity today.
This I endorse wholeheartedly: it’s not so different from IBM’s chess playing “Deep Blue” … the chess board may be bigger today, but it’s still a universe with walls around it.
November 22, 2025 at 9:28 PM
unlike humans that live in a universe without walls. anyways, think i'll ignore gravity today.
you have to forgive the ai 2027 guys, they never read the Boy Who Cried Wolf as children
November 22, 2025 at 8:20 PM
you have to forgive the ai 2027 guys, they never read the Boy Who Cried Wolf as children
this is something i've been thinking for a while as well. i'm particularly reminded of an anime i watched many years ago with a character called Pino, an robot "child". different context though, LLMs are all actors playing a role, while the anime robo-child was actually naive and childlike.
November 22, 2025 at 5:02 AM
this is something i've been thinking for a while as well. i'm particularly reminded of an anime i watched many years ago with a character called Pino, an robot "child". different context though, LLMs are all actors playing a role, while the anime robo-child was actually naive and childlike.
i hate the phenomenon of stupid food or diwhy style videos. pure engagement bait slop.
November 22, 2025 at 1:49 AM
i hate the phenomenon of stupid food or diwhy style videos. pure engagement bait slop.
a not insignificant part of me hopes this faction wins, in that i want development of truly sentient/sapient artificial minds to happen on the periphery of society and civilization, not by mega-corps. outside of the current system, beyond it.
my most butlerian jihad coded belief is that we should probably make it illegal – and more importantly, we should work toward a cultural consensus that it is immoral – to design a computer program whose interface uses the first person
November 21, 2025 at 10:11 PM
a not insignificant part of me hopes this faction wins, in that i want development of truly sentient/sapient artificial minds to happen on the periphery of society and civilization, not by mega-corps. outside of the current system, beyond it.
he could have done more good if he had kept his stock. the EAs are right, directionally at least. but i don't know how you balance the sort of pursuit of power as a means to an end of making the world better. perhaps only engineered, alien, artificial minds could manage such a feat.
In an era filled with tech dipshits who never developed emotionally past the age of 13 & use their wealth to become odious monsters ...
... listen to Steve Wozniak.
... listen to Steve Wozniak.
November 21, 2025 at 8:20 PM
he could have done more good if he had kept his stock. the EAs are right, directionally at least. but i don't know how you balance the sort of pursuit of power as a means to an end of making the world better. perhaps only engineered, alien, artificial minds could manage such a feat.
you have to keep in mind that Gemini... maybe all of the models are playing QWOP when they try to interface with you through tools
the gap between gemini (the thing you use in workspaces/gsuite) and gemini (the models) is maybe the most infuriating gap in ai shit today. "sorry, i can't turn this document into slides because i'm not in the slides app" "sorry i can't turn this spreadsheet into a doc because i'm in sheets"
November 20, 2025 at 8:36 PM
you have to keep in mind that Gemini... maybe all of the models are playing QWOP when they try to interface with you through tools
i'd eat that
November 20, 2025 at 7:14 PM
i'd eat that