5 Comments

This is fantastic, super informative, thanks Van!

Expand full comment
author

Thanks Jonathan! Appreciate you reading my work

Expand full comment

Love it Van. Hope all is well!

Expand full comment
author

Thanks Ben! Glad you liked it. Hope all is well with you

Expand full comment
Comment deleted
Expand full comment
author

Thanks Mike. Glad you found it insightful.

The main difference is that ChatGPT is trained on the open internet while BloombergGPT is trained on that + a bunch of proprietary data from Bloomberg including cleaned financial data, paywalled articles, etc. In the BloombergGPT academic paper, they provided evidence that BloombergGPT performed better on a number of financial tasks than a generalized model like ChatGPT. It's still early, but I think it'll be interesting to see over time how useful or not proprietary data ends up being in training models (especially domain specific ones).

I like the user interface for mid journey and have had more success with it! I wasn't super scientific in my choice, but it's worth giving it a go. It's also very easy to drop in reference images for your prompts. I'm not sure if you can do that on DALL-E. Maybe you can, I just don't know.

Expand full comment