We need constraints in Generative AI Models Now

Enough of unconstrained generative AI outputs. It often leads to wrong or harmful outputs. How to make LLM safe?

Time for constraints in models.

How can constraints help?

— — No unintended outputs

— — By controlling models, we control the right outputs

— — The regularization of gen AI can be done by governments by overlooking the inbuilt constraints.

— — Constraints are needed to avoid the wrong meaning of the same word.

— — For example, a prompt with Word Knife can trigger a wrong output.

— — A wrong output can trigger some wrong reactions by innocent people reading it.

— — Hence, governments must sanction AI companies to use constraints in their LLM models.

— — We right now aim only at Loss Minimization.

— — The feasible region is the entire internet, now in which gen AI finds solutions.

— The feasible region in an optimization problem is the region that are satisfied by constraints.

— Right now feasible region is the whole internet, including the dark Net.

— — We need to decrease the feasible region to a limited region from where LLMs can find solutions.

— — In layman’s terms, we need to set boundary points to gen AI feasible search regions.

— — We need it, as all internet data can’t be fed again in LLM, we have moved ahead.

— — So we tell LLM, what is right to search in the form of reducing the feasible region from the whole internet to several cricket grounds.

Thank you for reading.

Subscribe for updates.

References

(PDF) Unconstrained LLMs, How Long? Safe, secure, and trusted constraint-based LLMs