> That we haven't yet seen the successor is why I'm suspicious we're leveling out. Not clear if we're leveling out on what's possible, but price/performance ma

> That we haven't yet seen the successor is why I'm suspicious we're leveling out. Not clear if we're leveling out on what's possible, but price/performance may be increasing with model size. There's an economic limit. If the main way to fix hallucinations is to run the same query many times and look for consensus, that's going to be expensive.