I don't doubt that it's far easier to abuse traditional captcha systems, but I wonder how wide spread that is. A while ago I did a test with securimage and tensorflow/python/opencv/keras after I read a Medium post. While it could solve captchas with a little distortion when I added squiggles, dots, and more distortion it was unable to solve the captchas. I'm sure you could spend more time and create a system that can solve these captchas, I wonder how much effort some random spammer will put in to attack your blog. Yandex uses traditional captchas, and they don't seem to have any issues.