I'm assuming you're passing the -C parameter in the command line. This puts the fuzzer in a "crash exploration" mode, where it's expecting some already-known crashing inputs as the starting point.
If you want to fuzz the binary the usual way, just omit the -C parameter. If you eventually discover some cool crashing case and you want to examine how much control you have over the circumstances ot the crash, that's where -C comes in handy.
You should also really follow that core_pattern advice earlier in the screenshot. Without that change, you will end up missing crashes.
/mz