He critiques the traditional academic focus on catastrophic interference.
What are the real problems to be solved in continual learning? In my latest post, I tackle this question — reviewing where I think the field went astray in the past, how language models changed things, and where the real challenges remain. 1/2
Users praised the research examining plasticity and interference challenges in continual learning for LLMs as deeply thoughtful and interesting.
No Digg Deeper questions have been answered for this story yet.
Interesting essay
What are the real problems to be solved in continual learning? In my latest post, I tackle this question — reviewing where I think the field went astray in the past, how language models changed things, and where the real challenges remain. 1/2
What are the real problems to be solved in continual learning? In my latest post, I tackle this question — reviewing where I think the field went astray in the past, how language models changed things, and where the real challenges remain. 1/2
In short, I think the field focused too much on interference and negative effects of other tasks, and didn't focus enough on how learning could positively transfer in the future. Check out the post for the full story! https://infinitefaculty.substack.com/p/what-are-the-real-problems-of-continual
@AndrewLampinen Continual learning and alignment are the same issue: You have an ever changing frame of reference that needs to remain "stable/grounded" to something without knowing what that something is ahead of time. *in humans that grounding changes with time, ie epochs of mind.
@AndrewLampinen Deeply thoughtful and interesting read. ✨
@AndrewLampinen Very interesting read
@D2Chattoway Thanks!
@AndrewLampinen Out of curiosity, would you view projects like Hermes Agent as an instance of your first continual learning path: textual memory, retrieval, and later distillation? How large do you expect the gains from that approach to be on ProgramBench-like tasks (w/o reset constraint)?
@AndrewLampinen @willccbb
@dwarkesh_sp The AIBros REALLY need to understand this theory.
https://en.wikipedia.org/wiki/Dreyfus_model_of_skill_acquisition
@dwarkesh_sp Big Dwarkio 👹
