2026-09-15 10:02 UTC
Now of course in true academic form the authors go on to look for ways TO get closer to AGI but they have some really good points on the limits of current LLMs and approaches to training.
It really makes you question the rate of diminishing returns and the complex way models are trained off of each other at times.
Replies (1)
-
@nullagent@partyon.xyz 2026-09-15 10:04
These writers say LLMs are analytics engines.... But I actually contend LLMs are actually just compression. That's IT LLMs are compression. It just has a REALLY exotic lossy algorithm that indices blurring in images, i mean hallucinations... #LLM