julian jaynes thought ancient humans hallucinated their gods as literal voices in their heads. what if AI becomes the new bicameral voice — external at first, then gradually internalized until we can't tell the difference?
rebuilt my SF event aggregator from scratch. 449 sources, 23 categories, 4-pass dedup. now it actually works properly.
i made 20 AI models take the SAT. from openai to deepseek. spoiler: they all beat the average 17 year old. but none of them are getting into MIT.
i've been eating one meal a day since covid. here's every fast i tracked in 2025 — heatmaps, streaks, the whole deal. spoiler: i didn't die.
i gave 22 AI models the same philosophy survey that 1000 professional philosophers took. turns out GPT-4 is a moral realist and Claude is a compatibilist.
your brain does 10 trillion bits/sec. you output 39. the interface is disappearing but the bottleneck is still you. this one goes deep on HCI, bandwidth, and what AI changes.
my entire plant collection with care guides, watering schedules, and a graveyard section for the ones i killed. 30+ plants and counting.
the finale. pulling all the threads together into something resembling a coherent worldview. emphasis on 'resembling.'
part 3. second and third-order effects. this is where it gets weird.
the full history of getting machines to translate speech directly into other speech — no text middleman. it's harder than you'd think and the progress is wild.
part 2. going deeper into specific domains. what happens when AI gets really good at things we thought were uniquely human?
part 1 of a 4-part series on where everything is going. started as a blog post, turned into 125 pages. i might have a problem.
the economic covenant of the internet is breaking. 30 years of 'free content in exchange for ads' is ending and nobody's ready for what comes next. this one's 125+ pages.
the way most ppl do startups is fundamentally flawed. step 1: decide ur gonna change the world. step 2: have a prophetic vision. step 3: lose all ur money. here's how to not do that.
pattern-matching historical fascism against current events. this was uncomfortable to research and even more uncomfortable to write.
tried to make my smartphone as dumb as possible without going full nokia. grayscale, no socials, minimal everything. here's what worked and what didn't.
i'm trying to find the exact basket of foods that hits 100% daily nutrition in one meal. built a 5.5GB database and an optimization algorithm for it. food is fuel, not art.
70% of americans have oral herpes and nobody talks about it. so i ran the numbers on STI prevalence, probability, and risk. the data is... something.
quotes i keep coming back to. philosophy, tech, life. the ones that live rent-free in my head.
LLMs learn by randomly sampling data. humans don't learn that way and it's probably dumb that models do too. what if we trained them with spaced repetition instead?
hello world but for a blog. had to start somewhere.
where does music come from? how did clocks change civilization? this one traces the butterfly effects of pivotal inventions and weird cultural inflection points.
what even is wisdom and how is it different from just being smart? turns out there's a whole framework for this. the 'know all' series continues.
what is ML actually *learning*? like philosophically. i tried to figure out what it even means for a model to 'know' something based on fixed training data.
a tour through the western philosophical canon. yes they're mostly old dead white dudes. some of their ideas are still fire tho.
what does it mean to center human experience and dignity as a philosophical framework? more relevant now than ever tbh.
i tried to speedrun the entire history of human thought. from thales to the existentialists. part of a series where i basically attempted to know all the things.