Parasma puts living neurons on next-token prediction

Parasma’s result is an intriguing biological-compute milestone, but it remains a research demonstration—not a practical replacement for GPUs.

• –Neurons reportedly learned a basic prediction task from electrically encoded token sequences
• –The approach could offer radically lower energy use and native continual learning
• –Biological systems introduce difficult engineering problems, including noise, variability, observability, and reproducibility
• –Developers should expect experimental infrastructure rather than an accessible API or production platform
• –The bigger significance is testing whether living neural substrates can complement silicon for specialized workloads