a workspace in claude
anthropic found a "global workspace" in claude and open-sourced the lens. i ran it on a laptop in ten minutes. it works — and the model picks the shape of an answer before the answer.
personal corner of the internet
anthropic found a "global workspace" in claude and open-sourced the lens. i ran it on a laptop in ten minutes. it works — and the model picks the shape of an answer before the answer.
Meta's non-invasive brain-to-text hit 61% by reading people type. the people it's for can't type — and closing that gap costs signal on both axes at once.
what cognitive scientists have called memory for fifty years, what AI engineers are calling memory now, and how the two vocabularies line up.
what neurobiology knows about memory that AI agents forgot. on prediction error, bi-temporal modeling, and why most memory systems are junk drawers.
on imposter syndrome, positioning, and the gap between language knowledge and being good at your job.
first post. testing if this thing works.