← kwindla hultman kramer

I had so much fun talking to @jheitzeb during this recording

December 16, 2024

I had so much fun talking to @jheitzeb during this recording. We covered a lot of ground:
- Why I think time-to-first-token is even more important than token throughput for a lot of AI engineering. (And how we measure this for all the services we use at @trydaily.)
- How quickly voice-to-voice AI is being adopted in areas like customer support.
- The value of building as many integrations as possible to staying up to date with the crazy pace of change in AI today.
- Some speculation about what we'll all be excited about seeing from the big labs and open source hackers six to twelve months from now.

Joe Heitzeberg@jheitzeb

What if bandwidth was unlimited and latency wasn’t an issue? 🌐

How would that change the way we use AI?

In our latest “One-Shot” episode, guest speaker @kwindla explores whether, in such a world, you'd still run models locally or rely on an AI supercomputer with no cost or

Video from @jheitzeb's post