Lots of yummy quotables at Human[X] this week, Part IV

I'll end this series of yummy quotables with Mistral AI's founder and CEO Arthur Mensch. Arthur is very serious (in public), so here are two serious takes from his fireside chat:

1️⃣ "LLMs are not intrinsically safe or unsafe. They are used in systems and applications that need to be evaluated." This is very much in line with what a lot of AI researchers and experts have been saying, in particular Yann LeCun, with implications for regulations: regulate the applications, not the science. Of course, existing laws still need to be obeyed when doing the science 😇 .

2️⃣ LLMs are just "stateless functions" and therefore not very sticky. In other words, LLMs have no defensible moat unless they are integrated into a stateful system or application. There goes the "it's just a wrapper" myth! But it is ballsy for an LLM provider to say this, which shows that Mistral is not JUST an LLM provider.