The Complexity of Chat GPT and RLHF

I actually think RLHF and all these fine tuning techniques are reliant on this. It's hard to write RLHF right now. How do you share the state amongst so many different problems? Like I have to allow every node in that DAG to really easily share the model. And most of those nodes are going to be running at a different schedule because or they're going to be triggered by user actions. Yeah, like DAG frameworks and like execution engines are good, but I don't think there's a simple way to solve this or to do this.

Transcript

Play full episode

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.

No 1 podcast app

Steven

App Store

I tried everything and snipd is the no 1 app for podcasts if you like to remember things. Just tap your headphones three times and a snipped is created, transcribed, and saved to you library.

The game changer for learning from podcasts!

Nelson

App Store

I used to use a different app that was able to save excerpts from podcast and really enjoyed it. I could listen to the podcast and quickly save things that I wanted to come back to later. Snipd take this to a whole new level with AI integration, creating summaries of podcasts and summarizing the main takeaways from what I’ve saved and snipped. I really love how it helps me prioritize what podcast to listen to with it summaries & deep dives.