4 ms·
I agree, and I think we're a year or two away from a full end-to-end trained screen reader. The ground truth from existing systems would provide great training
by ctoth 3y ago
I agree, and I think we're a year or two away from a full end-to-end trained screen reader. The ground truth from existing systems would provide great training material.
As a technical blind person, my only concern is the inherent loss of privacy while sharing stuff with the big models.
- supriyo-biswas 3y agoThere are open source models such as https://github.com/THUDM/CogVLM https://github.com/THUDM/CogVLM and https://github.com/haotian-liu/LLaVA https://github.com/haotian-liu/LLaVA.