What would happen if we gave a single ai problem the compute currently used for millions of prompts?
Maybe I’m being naive, but whenever people discuss whether AI could make truly extraordinary scientific breakthroughs — curing cancer, for example — I get the impression that we may be looking at the problem from a very partial perspective.
We tend to think about the capabilities of an individual model answering an individual question, rather than about the sheer amount of AI “thinking” happening globally at any given moment.
Every second, LLMs are answering an enormous number of prompts from users all over the world. Collectively, that must require a staggering amount of compute.
So here’s my question: what would happen if, instead of using all that computational capacity to answer millions of unrelated questions simultaneously, we concentrated an equivalent amount of compute on a single scientific problem?
Suppose the question were something like: How do we cure a particular form of cancer?
Would concentrating that enormous amount of computation on one problem give an AI system radically greater capacity to search the literature, generate hypotheses, run simulations, test possible explanations, critique its own conclusions, and explore solution spaces?
Or is this based on a fundamental misunderstanding of how AI compute scales — i.e. you can’t simply turn millions of parallel LLM queries into one vastly more powerful act of “thought”?
I’m particularly interested in the distinction between more compute, more inference-time reasoning, and genuinely deeper scientific intelligence.