Can an agent solve an audio captcha that we generate?
Interested in agents and accessibility, I built an audio captcha generator of 100 character and math questions (which instead of asking for characters to type out asks how many of a single character are in the audio or a math question in audio form) - and tested if GPT4o could solve it if we let it use a trasnscription tool. Think how many r's are in strawberry but asked in audio form, or what is 1+52 in audio form.
Folks with visual impairments struggle with captchas
An audio captcha generator that can be used to verify users, testing if GPT4o could solve it if we let it use a trasnscription tool
Explored accessibility and agent use cases