Replit AI Podcast cover image

03: The Next Generation of LLMs with Jonathan Frankle of MosaicML

Replit AI Podcast

00:00

How to Match Chat GPT Across the Board

The chat model is really good, but it's not commercially usable due to the origin of the data. I think at this point, we're now seeing the result of a lot of efforts that probably started a few months ago to train llama quality models. We just need to be willing to be a little bit creative and not just train llama again for the end time. People should be willing to do weird stuff. And we can afford to do that on instruct tuning and fine tuning. It's hard to do it for base models. Oh, I completely disagree. Do it on a 300 million parameter model and work your way up. At least convince yourself that you are doing something different

Transcript
Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
Get the app