81·
1 year agoWriting good alt text is more than just identifying the objects in an image. It requires some understanding of the context around why the image is being included. Harvard provides some great advice for humans to do this well at https://accessibility.huit.harvard.edu/describe-content-images.
Unless Elon puts Tesla’s FSD team on this problem, it will be many years before AI can do this well. With the FSD team on it, it would only take a decade.
@shapis Almost 20 years ago, I followed Lawrence Lessig’s RSS feed. He made a request for software that could be used to advance slides on a remote computer. I knew AppleScript fairly well and thought, “how hard can that be?”. I wrote a one script that would “listen” for the text “Next Slide” in iChat and then try to advance whatever was open in PowerPoint. I wrote another script with a basic UI so the presenter could easily “type” Next Slide while presenting. It was basic, but it worked. I think I shared the code with an MIT license. Even though the code was free and Dr. Lessig already agreed to meet with a class about IP Law at the university I was working for at the time, he also contributed $50 to my project. He could have just downloaded the scipts and used them without paying anything, but that simple act changed my life. I realized that some people who could afford it would pay for code I even when I was giving away. Most people don’t, but enough do that I’ve been able to continue contributing my code, helping to fix bugs in other people’s code and sponsoring other projects today.
https://archives.lessig.org/indexb00c.html?p=2897