@sun @leyonhjelm @mischievoustomato @s8n there are very exciting things from the past year or two but they aren't from GAFAM.

they reverse engineered eligibility traces (tl;dr approximate the derivative of backprop-through-time down to a single protein marker), made a basic immitation of top-down learning signals (training the learning network to train the bigger one how to walk)

assuming all the spike shit have from last month wasn't made up in the journals it really is just a matter of taking the parts and figuring out the right way to jigsaw them together (i say this as though it's not still an immense pain in the ass)