Use cheaper, faster AI models where the output is good enough

Phil Carter… it's controversial because at this point the best open AI and anthropic models are just so powerful that it can be easy to just say, "Well, why would we try anything else?" But Gamma is a great example of a company that when they were first getting started, they famously had this viral tweet in March 2023 when they launched AI-powered presentations that finally altered the trajectory of their business and six months later they were profitable.

Phil CarterAnd there are a number of reasons why they got to profitability so quickly, but one of them was they were very savvy in the underlying LLMs they used to power the early versions of their AI powered presentations product, which is not to say they weren't ever using the most powerful models, but in some cases what they found was using some of these longer tail models, not only was the performance good enough in terms of the quality of the output that they were delivering to their users, but also the speed of the product was significantly faster and the cost to the business of the compute power needed to serve these AI powered features was significantly lower, which as a startup that was fighting for its survival in early 2023 was really important.

Phil CarterAnd so it's just a lesson that at this point we're all like waiting with bated breath for the next release of the hot new open AI or anthropic model. And that's great. The frontier of what these models can do is so fascinating and compelling, but don't forget that there's this long tail of thousands and thousands of other models out there. And sometimes those models tailored to your use case can be the optimal solution.