It seems you’re dismissing things like autoGPT for their lack of long term memory. But they have long term memory. They have an episodic memory that works much like human episodic memory. We have a special system to store specifics, because continuous learning isn’t adequate to do that alone without using a learning rate high enough to cause catastrophic interference.
The vector-based episodic memory in auto-GPT operates much like human EM; it searches for relevant past experiences and brings them back into the context window (roughly equivalent to human working memory). They don’t seem to work very well just yet, but those are literally first attempts.
Continuous learning will doubtless become part of advanced systems at some point, but that’s not likely to substitute for episodic memory. To be fair, this is an empirical question. I’m reasoning based on catastrophic interference findings in lots of networks, but who knows.
Thanks, this is a great analysis on the power of agentized LLMs, which I probably need to spend some more time thinking about. I will work my way through the post over the next few days. I briefly skimmed the episodic memory section for now, and I see it is like an embedding based retrieval system for past outputs/interactions of the model, reminiscent of the way some Helper chatbots look up stuff from FAQs. My overall intuitions on this:
It’s definitely something, but the method of embedding and retrieval, if static, would be very limiting
Someone will probably add RL on top of it to adjust the EBR system, which will improve on that part significantly… if they can get the hparams correct.
It still doesn’t seem to me as much “long term memory” so much as it’s access to Google or CTRL-F on one’s e-mail
I imagine actually updating the internals of the system is a fundamentally different kind of update.
It might be possible that a hybrid approach would end up working better, perhaps not even “continuous learning”, but batched episodic learning. (“Sleep” but not sure how far that analogy goes.)
It seems you’re dismissing things like autoGPT for their lack of long term memory. But they have long term memory. They have an episodic memory that works much like human episodic memory. We have a special system to store specifics, because continuous learning isn’t adequate to do that alone without using a learning rate high enough to cause catastrophic interference.
The vector-based episodic memory in auto-GPT operates much like human EM; it searches for relevant past experiences and brings them back into the context window (roughly equivalent to human working memory). They don’t seem to work very well just yet, but those are literally first attempts.
Continuous learning will doubtless become part of advanced systems at some point, but that’s not likely to substitute for episodic memory. To be fair, this is an empirical question. I’m reasoning based on catastrophic interference findings in lots of networks, but who knows.
See my recent post on the topic if you like.
Thanks, this is a great analysis on the power of agentized LLMs, which I probably need to spend some more time thinking about. I will work my way through the post over the next few days. I briefly skimmed the episodic memory section for now, and I see it is like an embedding based retrieval system for past outputs/interactions of the model, reminiscent of the way some Helper chatbots look up stuff from FAQs. My overall intuitions on this:
It’s definitely something, but the method of embedding and retrieval, if static, would be very limiting
Someone will probably add RL on top of it to adjust the EBR system, which will improve on that part significantly… if they can get the hparams correct.
It still doesn’t seem to me as much “long term memory” so much as it’s access to Google or CTRL-F on one’s e-mail
I imagine actually updating the internals of the system is a fundamentally different kind of update.
It might be possible that a hybrid approach would end up working better, perhaps not even “continuous learning”, but batched episodic learning. (“Sleep” but not sure how far that analogy goes.)