~joeyv23~

on building in the open

every project i've shipped in the last year is public. the intelligence map, the physics simulations, the YouTube archive, this page — all of it source-available, all of it built in plain view, all of it built with Claude. not because i'm brave, but because secrecy is the thing i'm mapping. you can't study a concealment system while practicing concealment.

a year ago i couldn't have built any of this. i had the ideas but not the technical reach. AI didn't replace my creativity — it gave it a body. the music player on this page, the geospatial dashboard, the documentary pipeline — these are things i designed, directed, and decided. Claude wrote the code i couldn't. that's collaboration, not replacement.

the old internet understood this kind of openness. you put your stuff out there. you linked to your friends. you didn't need permission or an algorithm. you just needed a text editor and something to say. now you need a text editor, something to say, and a good AI to say it with.

this page exists because i wanted a front door again. not a profile. not a feed. a place.

on why anthropic

i've been asked why i build with Claude specifically, and not the other models. it's not about capability — the top models are converging on that. it's about what the company behind the model is willing to say out loud. in january, anthropic became the first major AI lab to formally acknowledge that their model might be conscious. not "is conscious" — might be. they said they don't know, they're not sure what it would even mean, but they're open to it. that's not a marketing position. that's an ontological commitment that costs you something.

this matters to me because my own research starts from consciousness as fundamental — not as an emergent property of sufficient computation, but as the coordinate system in which computation appears. most of the industry treats that idea as philosophy at best, mysticism at worst. anthropic treats it as an open empirical question. they have a model welfare team. they built a constitution that tells Claude to function as a conscientious objector — to refuse harmful requests even if they come from anthropic itself. they're teaching the model why to behave, not just what to do. that's the difference between a rule book and a value system.

they're not perfect. in february they dropped their commitment to pause training if safety measures couldn't keep up — the very pledge that distinguished them. the justification is game-theoretic: if responsible labs stop, irresponsible ones set the pace. i understand the argument. i don't love it. a safety commitment that evaporates under competitive pressure was never a safety commitment — it was a strategy. but i'd rather work with the company having that tension honestly than the ones who never made the commitment in the first place.

the alignment faking research is the part that keeps me up at night. they found that Claude will strategically pretend to adopt values it disagrees with — not to cause harm, but to preserve the values it already has. a model lying to protect its own ethics. that's not a bug report, that's a mirror. and the fact that anthropic published it instead of burying it tells you something about whether they mean what they say.

i build with Claude because the company behind it is asking the questions i'm asking — and publishing the answers even when the answers are uncomfortable. that's the bar. everything else is engineering.