friendica (DFRN) - Link to source
LLMs and accessibility

So, I keep seeing people using LLMs to generate alt-text for images. From an accessibility standpoint, I guess this is better than no alt-text at all, but...

I notice that the descriptions they generate are... often technically accurate... I guess? It's just that they really seem to suck at infering from context what parts of the image are noteworthy and what aren't.

in reply to Jonathan Lamothe

mastodon - Link to source

Ryan Robinson

re: LLMs and accessibility
To be fair a lot of humans are also very bad at writing alt text, if they try at all. I just hope the LLM assisted ones are a lot more coming from an attitude of "I can start from the generated text then learn to do it better" or at least "this is all I have time for and it's better than nothing" rather than "I don't have to think about this anymore because the computer will do it."

This website uses cookies. If you continue browsing this website, you agree to the usage of cookies.

⇧