Are We Looking for AI Control in the Wrong Place?
Billions of dollars. The sharpest minds in computing. Entire industries built around one question: How do we make an LLM do the right thing—predictably, repeatedly, every time? More training. Better prompting. Fine-tuning. Guardrails inside the model. More evaluation after the model has already made the choice. But what if the industry is looking in the wrong place? What if the answer was never…
The quest to mold large language models into obedient, predictable tools has consumed vast sums of money and the brightest minds in computing. Countless industries have been dedicated to refining these probabilistic models through more training, superior prompting, fine-tuning, and intricate guardrails. Yet, it's conceivable that the industry has been searching in the wrong place.
Perhaps the solution does not lie in forcing a probabilistic model to conform, but in eliminating its authority over decisions that should not be left to chance.
Artificial Deterministic Intelligence (ADI) poses a provocative question: Why is the model being asked to decide on permission, authority, policy, safety limits, or release conditions to begin with? Why not allow the model to reason freely, while maintaining control outside its system? This approach is deterministic, auditable, and ultimately, owned by people rather than the model itself. While this method does not guarantee an infallible LLM, it does shift the locus of decision-making power back to human hands.
However, this paradigm does not render the LLM universally correct. It merely changes who holds the power to make crucial decisions. As the industry grapples with the strongest technical objections, one must ponder: Which decisions do you believe must remain firmly within the confines of model inference, and why?
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
