AI model

“It is possible to build a model, but I think it's also possible to fine tune a model which will get you 90% of the way there.”

“The underlying intelligence and model is being commoditized, right, then where is differentiation going to come? How do you build defensibility as a founder? If it's not the model, which for 90% of folks it isn't, then it is either the data you're bringing to …”

“We've seen this firsthand as a few of our ideas have been basically one-shotted by newer models. And basically our building of it doesn't make sense anymore because a new model comes with a sub feature that solves that.”

“Make sure you have an amazing end-user product or value proposition but be really flexible about how you build your back-end architecture, because it will be obsolete in 18 months. Actually, your front-end, your Product may also be obsolete in 18 months”

“However, humanness, which is what, how I behave in a certain context scenario, how I talk, how I might interrupt another person, how I might express my opinion, that does not require large, large, large models. It actually requires the opposite. Smaller models…”

“I think in, in the history of technology, it's probably the first time that we're seeing a tool or a product that that's behaving in a way that we cannot predict in a way, right? So it's like, you cannot expect a certain thing that come out of the model, right…”

“And you're seeing now too, with some of the stuff Claude is putting out, or Anthropic rather, security holes are going to be easier to find with the more intelligent models.”