Comment by mastazi
17 hours ago
So "decision model" is the new cool term for classifiers? You know, that thing that already existed decades ago
17 hours ago
So "decision model" is the new cool term for classifiers? You know, that thing that already existed decades ago
decades ago you would train a classifier over a fixed set of classes. decision models are flexible regarding the possible outputs they can produce. also, you get text and images in input
But they’re only flexible because they can hallucinate an answer for anything. It’s like if you asked Gary Oldman to make all your decisions for you. He could do it, but he might have to make up answers when you get to more difficult topics
But he would also tell you how confident he was in his guesses
On a side note - this would make a brilliant TV show. Gary Oldman for a day making life changing decisions then we come back one year later to see how it all went. A bit like Grand Designs.
Modern version of « The Dice Man »
If you think “bullet” is the cool new name for “arrow” then sure.
Decision models can perform zero-shot classification over almost unlimited text-input domains. This was science fiction decades ago.
A coin flip can make infinite decisions on all possible input taxonomies.
Zero-shot is indeed very cool. And it's super useful/useful for the developer masses that don't know, care, or work pressures don't allow, for proper evaluation and calibration.
But it would be good that we don't over-hype these things.
A biased coin flip with accuracy on my domain is what I all care.
The difference is in the training, or more accurately what data the model is trained on.