LLM-QP

Most LLM optimization talk is about making models smarter. LLM-QP is about something quieter: the same correct answer, reached at a fraction of the compute. If the output is identical, the gap was never intelligence.

A model that gets the right answer at 100x the necessary cost isn't underpowered. It's over-provisioned. LLM-QP is the research thread on closing that gap.

The interesting framing in LLM-QP: if two runs produce the same correct output and one costs 100x more, you don't have an intelligence problem. You have an infrastructure problem. That reframe changes what you optimize for.

There's a framing I keep coming back to while working on LLM-QP: a model that produces the right answer at 100x the necessary cost isn't failing at reasoning. It's failing at resource allocation. Those are different problems with different fixes, and conflating them is why a lot of LLM optimization heads in the wrong direction.

A correct answer at 100x the necessary cost isn't an intelligence failure. It's an infrastructure one. That's the framing behind LLM-QP. https://www.jakelawrence.xyz/research/llm-qp

Correct answer, 100x the necessary compute. That's not an intelligence gap, it's an infrastructure one. The research thread is at LLM-QP.