AaronThey memorize facts and thoughts and data, not just reasoning and language patterns. The vast majority of the enormous number of parameters in these models goes towards memorization of the entire internet. The form this memory is stored in is "squishy" -- pushing on it to imprint something new deforms the metaphorical putty it is crafted from and erases other things. You have to press in exactly the right shape from every direction at once (training on the whole corpus) if you want it to truly hold its shape. That's not how human brains work. We learn incrementally, continuously, piecemeal. The tech behind LLMs can't be retrofitted for that. It has to be something new, something more brain-like.
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.