increasingly clear that 'data' companies (rl envs, human / synthetic data ...) need to train their own models, on their own data, to solve real problems in any vertical. Then ofc you can sell the curriculum The most successful data companies will look a lot like a neolab!