Cohere presents TIPSv2 vision-language pretraining method

Talk covers TIPSv2, an enhanced vision-language pretraining approach with spatial awareness demos. Key findings include patch-text alignment improvements and a distillation recipe upgrade to iBOT++.