prediction - data ends up costing waaay more than compute to train ai models. seriously, the more I dig into it there’s just so much untapped potential in *existing* models that could be unlocked with the right context

but the problem is model creators don’t have access to niche data or the data is “unreadable”

right now googles paying $200m at the top end of the market for high quality data but that’s just for generalised models

once this shit starts getting agentic, it’s going to need data to provide context on an infinitive number of potential scenarios before picking one and executing - that data either makes or breaks the quality of the users experience. meanwhile compute eventually gets commoditised.

we’re nowhere near this trend flipping yet but worth paying attention to data aggregation, sorting, synthetic generation & RL environment companies.