3 ms·
My hunch: It's better that we don't hard-code hard-won specialised knowledge. It's better that this tech can learn from more fundamental foundational principles
by igravious 2y ago
My hunch: It's better that we don't hard-code hard-won specialised knowledge. It's better that this tech can learn from more fundamental foundational principles that can be leveraged to apply to many many specialised domains.
- igorkraw 2y agoNah, it's much simpler, the models aren't reliably able to recall the correct rule from memory - it's im the training set for sure. This is another specialized synthetic data generation pipeline for a curriculum for one particular algorithm cluster to be encoded into the weights, not more not less. They even mention quality control still beim important
- InvidFlower 2y agoWell, not sure if that part matters as much (from first principles). But the more important part being that RL lets a model figure out which methods are effective for it. Most of the time it probably has the tools already from pre-training, but doesn't "make the connection" to use them (or at least not often enough).
- samstave 2y agoF_yes -- The thing to take here is that this should be a function_callable *feature* of a bot. Basically, when architecting a persona ; "USE THESE RULES OF ENGAGEMENT" Also -- where my CheckListManifesto folks at -- While we are building Patterns/Personas/Purgatories for our bots... We need to be able to reference a central CODEX of : "Do this task but imbue yourself with (XYZ) name places" -- AND LEARN FROM THE OTHERS (so maybe a task marketplace of AI persona action?) @ callable in an IDE