There’s a level of unhappiness and distrust of the frontier labs from CEOs
There needs to be an application layer on top of LLMs (e.g. “harness”, Palantir Ontology)
This application layer prevents the LLMs from learning your business from your data
“Alpha” is business differentiation (ability to outperform the market)
He questions why the frontier model labs are charging by tokens and not outcomes (questions the entire AI business model)
He questions “the true cost” of AI outputs
He claims that CEOs are now concerned about frontier labs entering the business of the customers - brings up an interesting misunderstanding of how interacting with LLMs works (“we’re safe, it’s deployed in our VPC”)