With all due respect, this is pure cope. There's no magical way to train a smaller model. The scaling laws have solid mathematical underpinning that you cannot wish away. Do you think that the frontier labs have not attempted to get to a more efficient architecture?
If the goal is to be frugal, we can post RL train off an open source base model. Has Zoho published anything? Why is the *most obvious* low hanging fruit not tackled?
I agree with Brian here.
The chat interface is such a bad implementation in this usecase because it's tiring for the user to type everything in a text input we need to reimagine the input -> response layer
@benhylak I’m not saying Airbnb’s current interface is optimal for AI. It isn’t. The future interface should be conversational, but visually richer than a chatbot. It’s a hard problem, and we haven’t solved it yet.
@Sankew06 Most probably they’re going to add a triangular folding placeholder on the edges. When unfolded, it fills the corners making it look flat and when folded, the plate folds inwards