Ah the irony of defending an AI with a method that an AI should be able to solve. So the captcha solver will own the LLM, it's evolution baby!
Leaning towards the former more.
LLM websites need more profit and more data, so all they have to do throw in new training questions to force their already paying customers to do more work for them, before they can do their own work! I can already feel microsofts drool on my neck
OpenAI farmed pretty much all the quality text out there for GPT-4. They did GPT-4o built natively for multimodal, likely because they ran out of text and need to farm videos & audio. Captcha data is pretty valuable too if only to beat captchas.