Hacker News Reader

Comment by igravious | Hacker News Reader

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.

View on HN

Nah, 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

Well, 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).

F_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

Reply on news.ycombinator.com