And again, the world shows that our accessibility needs don't matter, but when loser nerds need them, The Web Is Being Made Accessible for AI, Not People https://www.techpolicy.press/the-web-is-being-made-accessible-for-ai-not-people/ #Accessibility #A11y #AI #LLM
@WeirdWriter@caneandable.social the silver lining is that the MCP protocol is relatively simple and you could probably use it to make a more accessible UI for some services, or build some tool that lets you talk to MCP servers directly and manually use the functions it exposes, pretending to be an AI agent.
Not that the “AI friendly web” is a replacement to following WCAG guidelines, it's truly infuriating that that's an afterthought.
@WeirdWriter@caneandable.social
> “Plaintext documentation optimized for a language model is not the same as well-structured HTML for a screen reader.”
The frustrating thing is that it wouldn't be difficult to provide structured HTML in such manner where a preprocessing function could extract the main content and plaintext trivially.
As far as I can tell, plaintext isn't necessarily the best format to get unless you literally just dump it all into the model's context. If you have the semantic structural information, you can divide it into logical chunks and build a database which a model can look up concepts in that doesn't require a huge context with massive compute resource requirements. You could do the same for general search too.
Either way that would align the AI friendliness with accessibility much more.