[teknopurge]

>> Almost nobody will also train their own models ...

If the value is in better training data then everyone will train their own models. We should have a goal where we have domain-specific models that make us more efficient, but also constantly train a context-aware model based on individuals. Model composition is the future.

[tomp]

> everyone will fine-tune their own models

[teknopurge]

individuals / companies that cannot afford their own workflows will fine-tune models with heavy preexisting biases. The upper-class of the market will train their own models, producing advantaged outcomes. Data is the moat.

[emadm]

Thus there is a market on creating open, auditable models of every media type, with sectoral and national variants.

Models are like talented grads that occasionally go off their meds, proprietary black box models are like hiring them from McKinsey, fine tuning your own ones that you control is like hiring them

All govts, regulated industries etc will use open auditable models so much bigger TAM

[felipeodonnell]

Considering proprietary black box models are going to be present in some form, is the difference between having AI lead to huge societal benefit or big brother down to whether data is open and provided by governments or closed in big tech?

[emadm]

Yes if there aren’t good open models available then it’s massively centralising and will lead to a massive panopticon