3 ms·
I'd recommend looking up model grounding by multi-modal training. Seemingly models improve as you add more modes.
by pixl97 3y ago
I'd recommend looking up model grounding by multi-modal training. Seemingly models improve as you add more modes.
- daveguy 3y agoThe study did include a multimodal model. Apparently it doesn't improve abstract reasoning capability, because according to the article the multimodal gpt4 did just as dismally as the text-only gpt4. This was surprising to me, as I would have expected an improvement with a model that did include spatial relationships.