Intelligence is nothing without experience, so we should probably aim to define certain level of intelligence applied to a certain knowledge base.
Further, there is depth and breadth of this knowledge which may define a type (specialist vs generalist) and level (shallow vs deep) of expertise, respectively.
I’d expect the nature of the data used to train a model would be reflected in the cost per unit of work: more proprietary the data, more expensive it would presumably be.
But assuming for a moment we can use IQ as the metric. I wonder what HN thinks what’s the level of IQ-equivalence of the state of art GPT-4o model.
Wonder if mere IQ is the right metric.
Intelligence is nothing without experience, so we should probably aim to define certain level of intelligence applied to a certain knowledge base.
Further, there is depth and breadth of this knowledge which may define a type (specialist vs generalist) and level (shallow vs deep) of expertise, respectively.
I’d expect the nature of the data used to train a model would be reflected in the cost per unit of work: more proprietary the data, more expensive it would presumably be.
But assuming for a moment we can use IQ as the metric. I wonder what HN thinks what’s the level of IQ-equivalence of the state of art GPT-4o model.