@johnbender Yeah I know what you're saying, and I suppose for specific tasks (e.g. writing emails or marketing materials) the fine tuning corpus is a set of human written things for which it will probably be average at, unless the base training helps somehow
@hankgreen I think it comes down to the sparseness of training data relative to the lived experience of a human. The model doesn't get to see all the minute details that would make a realistic scene, just sporadic snapshots of life that it then interpolates into images that seem dreamlike
@OtterMarket You can start with the fed having to raise rates to combat inflation and investors demanding higher yields in general. This means future cash flows are discounted more, lowering the value of equity. Also higher rates means less cheap money to spend, so less demand for goods.
@andrewhuang I think youβre nitpicking semantics here. What people are hearing when they say that is that they can hear how perfectly quantized much of electronic music is, vs a band where each bar has human variation in timing. Itβs a different feeling