3 ms·
It appears like the 'self-improving' here just means modifying the agent's prompt/context? And not actually changing any of the weights/architecture of a model.
by sva_ 7mo ago
It appears like the 'self-improving' here just means modifying the agent's prompt/context? And not actually changing any of the weights/architecture of a model. I feel like this kind of self-improvement has some hard limits on how much it can improve.
- internet101010 7mo agoDefinitely isn't perfect and has limitations, but if the goal of predictable outcomes in a dynamic environment at scale it's more feasible than creating fine tuned models for every little thing and allows for context-based model performance benchmarking.