3 ms·
This is one of the reasons ML benchmarks are most useful soon after their creation. Once they become a target, they cease to meaningfully measure quality. The
by huevosabio 2y ago
This is one of the reasons ML benchmarks are most useful soon after their creation. Once they become a target, they cease to meaningfully measure quality.
The corollary for the Soviet shoe factory is you need incentives that propagate to guide production. You need the market to guide you in all the little details and you have to discover little by little what it is actually demanding.
For ML, we need something similar if not the same.
- mempko 2y agoThe market is notorious for creating a lot of distortions. I don't think the answer is 'let the market guide you'. Externalities are a known but huge problem with markets as evidenced by the insane amount of pollution we produced.
- kmoser 2y agoWhen the market isn't demanding what you're supplying, call in the marketing and advertising departments to get your customers to value what you're producing, rather than acquiescing to your customers' demands.