Chad@chadbailey59August 20, 2024
Two things I did not expect when we started seriously building voice AI infrastructure and tooling a little over a year ago:
1. Each step function improvement in LLM capabilities has come with more "personality." This maybe isn't apparent if you just use basic prompts, or prompts focused only on tasks or outcomes. But personality really shows through as soon as you start looking for ways to encourage it.
If we think of personality as just a category of capability, this makes sense. But it wasn't obvious to me that we'd see more *interesting* behavior from models along non-utilitarian dimensions, as quickly as we have.
The progress in voice quality for text-to-speech makes this ability of LLMs to perform as characters even more impactful. In the video below, the voice is one of the @cartesia_ai standard voices.
2. You don't have to call it "prompt engineering" — some people love that term, some people hate it — but it's a real thing.
You get better at using LLMs the more you use them. There are a bunch of things that go into developing an LLM skillset: figuring out what the models are good at and not good at; putting tooling of various kinds around LLMs; keeping up with the breakneck rate of progress; etc.
But the single biggest lever is the prompt. Prompt skills unlock huge surface areas of LLM capability.
Watching @chadbailey59 and @JonPTaylor one up each other adding new prompts to the Daily Bots playground demo has been a lot, lot, lot of fun.
Go try it yourself: https://t.co/yWEy4JkZWF
Or fork it and build something different: https://t.co/bAZ9NHzBnh
Chad@chadbailey59@kwindla @ProductHunt Meet... John. Just an average, normal human person
