Failing at the first hurdle
I am concerned. Not about Large Language Models (LLM)s doing me out of a job, but about the people who make money out of them not making their User Interfaces (UI)s accessible.
Not about their ethical use, though there is plenty to be concerned about, but about the accessibility and usability of their UI’s.
The biases inherent within LLM’s is writ large upon the UI’s provided to make use of them.
A cursory review of the ChatGPT UI shows it is broken for many people:
Keyboard operability desert
The Chat history is largely not navigable or operable using the keyboard as the links are not focusable. For example, the Custom Listbox Example “link” is an <a> element without a href or tabindex, thus it cannot be navigated to or operated using a keyboard.
<a class=”flex py-3 …”>
<svg stroke=”currentColor” fill=”none” stroke-width=”2″ …>
<div class=”flex-1…”>Custom Listbox Example
Feast of unlabelled buttons
If you have good eyesight you can sort of work out what the various buttons in the UI do, but if you [...]