@bignose @essjax I can confirm this.For starters, if you just toss a picture at an LLM, especially without a specific pr...

@bignose @essjax I can confirm this.For starters, if you just toss a picture at an LLM, especially without a specific prompt (and alt bots don't get specific prompts), the LLM will write a generic image description. Not one specifically for the context in which the image will be posted. It won't take the expected audience into consideration either. It literally won't know the context or the audience.An LLM doesn't know any rules and guidelines for good image descriptions and alt-texts, especially not those specific to the Fediverse. So it won't follow any of them. (Granted, most human Fediverse users don't know any of them either. They just wing it without even knowing that they're winging it.)And then there's the widespread attitude that LLMs can describe any image more precisely and accurately than any human. Whoever says this has probably only ever posted stuff like real-life cat photos. But what may work for cat photos (LLMs can even fail at these) certainly will not work for extre...

Read Original

Related