SI (Stanford Indians) just solved continual learning in LLM’s. 4200x more efficient, 27x faster.https://arxiv.org/abs/2610.21375Tick tock, luddites.
>>110260552use case for 12 token answers?
>>110260552> Using only 16 memory tokens, MEMORILLA achieves the best average performance against baselines including Mem0 and MemoRAG across benchmarks spanning personalization, long-document and knowledge-intensive question answering. Notably, on FactKG, it reaches 84.33% accuracy versus 80.43% with full-context inference, while using 108× fewer tokens to represent the source documents.
>>110260552>Now you can pay less money to hallucinate even more!Thanks, jeet
>>110260755> to hallucinate even moreumm.. but that’s.. literally wrong lol. Are you like.. retarded? lmao. It’s literally in the paper. Like why would i post it if they said it hallucinates more… what a dumb luddite. Let me guess, some permutation of being a teenager/tranny/codecel
>>110260552You don't even know what you're talking about, you fucking faggot.The guy is talking about yet another agent memory system, nothing to do with "learning in LLMs". That term is never used in such context that way.
>>110260552Total VibeGOD victory
>>110260552a headline like this comes up twice every weekall frauds
Cool, that's why im not shelling out money for ai hardware lmao, the llm space is continually changing Poor cuck oldman and the rest of the compute grubbers are sitting on old stock already geeeg
>>110260552continual learning in LLMs has been "solved" like a dozen times by now.
>>110260827> nothing to do with "learning in LLMs"…> Overall, Memorilla aids frozen LLMs in accumulating and using knowledge effectively across long-context interactions and takes a step towards building continually learning AI systems.awkward..
so maybe this is actually the key? fly brains?maybe we only need to have a memory capacity of like 256 bytes?
>>110260552Awesome.What has the "stanfort whites" group been up to?Useless?
