3 ms·
I'm surprised nobody has mentioned the connection to AI. Many models need "positional encodings" as part of their input in order to to understand the spatial or
by modeless 4y ago
I'm surprised nobody has mentioned the connection to AI. Many models need "positional encodings" as part of their input in order to to understand the spatial or temporal relationship between different input tokens. It's not at all surprising to me that the brain would have an equivalent.
It's not always clear what form these encodings should take. If we could figure out the actual encodings used by the brain, I bet we would find that they have big advantages over the ones we're currently using.
- carbocation 4y agoCame here looking for this comment. Even as someone who doesn't have feelings one way or another about any brain-neural network similarity, this one struck me the same way. Thanks for mentioning it.
- dogcomplex 4y agoOr the inverse - like "garbage" dna, our minds encode memories with "inefficient" baggage stamped to them... which are very likely how we tie emotion to memory. Perhaps a mind stamping with simple math could be much more "efficient".