@nate_512Naomi Bashkansky left OpenAI and joined Conduit where the goal is non-invasive mind reading, no implants, no surgery, just training models to pull thoughts straight through the skull. Right now their AI hits about 45% accuracy on a stranger’s brain, and they’re gathering data in an SF basement running entirely on batteries because wall power creates too much electrical noise. They’ve already banked 10,000 hours of brain data. Bashkansky calls it the “GPT-2 era of mind reading”, early, clunky, but scaling fast. For context, Whisper needed 680,000 hours to nail speech, so there’s a clear path to pushing that 45% way up. The timeline they’re eyeing: devs wearing thought-streaming headbands by 2027, and by 2035 AI feeling less like a tool and more like a limb
View original post

















