Hung thinks the bad rap is unfair. "You don't have to train your own models to be different, and you can be a GPT wrapper without having all of the bad marketing names to it." He reframes a model call as just "a function in the code that allows your platform to be intelligent."
More from this episode
Hung says AIducation does train its own models, but argues training isn't what makes you different. He draws the analogy that companies like Oracle and Google have customer-facing software that, at its heart, is "just databases" — so instead of thinking about it as a GPT wrapper, "you can think about it as a function in the code that allows your platform to be intelligent." Training mattered for his team specifically because they wanted a model that could converse according to the standards of pedagogy in Vietnam and solve math, English, physics and chemistry problems correctly — but "for most founders and most companies, I don't think that would be needed." He cites cow AI as an example that succeeded with "smart tricks and smart engineering," not heavy training.