Hacker News Reader

Comment by JSR_FDED | Hacker News Reader

Very interesting. They test Kimi K3 and Fable on a set of approx 1000 tasks grouped into 5 areas (SWE, Legal, etc).

They put a router model in front that predicts whether Kimi or Fable is going to give a better cost for a correct result. (They believe that ultimately such a router model should be continuously trained on your own workloads so it makes the best decisions for you).

Their router chose Kimi the majority of the time (72% in one category, all the way to 96% in another category), leading to cost savings in every category (from 1.5x to 50x depending).

View on HN

> Oracle routing is a method for measuring the best theoretical performance by running the task through each model and then picking the cheapest correct option (the cost/performance ceiling).

Their "router" is an oracle reference point where they choose the lower cost model after running both and therefore knowing who passed the test. The cost savings part is only Fireworks theorizing what would happen if an equivalent predicting router exists. That's a big if.

The way they published this is baffling to me... surely you can try to implement some router and then see how well it does. Using an Oracles makes the whole writeup so much less interesting.

But then that becomes an article about the performance of your draft router. This one is about the fact that there's this level of optimization potential.

> running the task through each model and then picking the cheapest correct option

What system knows what the correct option is and how does it know it?

Certainly sounds like a "P vs NP" style conjecture that shouldn't be possible in practice, save for certain generalizations, such as "this is a cybersecurity task, we know Fable will refuse (and score zero), so we just route to K3".

I agree: it feels to me that judging this implies knowing whether there is a solution at all (or a solution available per model, example: whether either model will answer it given known guardrails), which is as powerful as answering the question in the first place (the router can answer the decision problem, which can polynomially be transformed into getting a specific solution).

It reminds me of someone I met at a poster session who had an incredible project: his ai could include a confidence score with its answer, and he had evidence that x% confident answers were in fact correct x% of the time. That also gave me the gut feeling of that being impossible (in the general case) as it implies more powerful capability than the ai answering the question, in an oracle like way.

P vs NP?! What?

There are a bunch of routers like this eg https://openrouter.ai/openrouter/auto

Reply on news.ycombinator.com