How can this async url response checker be cleaner/faster?

Overall, your code looks pretty decent. The functions are doing what they should be (create_task shouldn't be running the tasks as well), coroutines are gathered after aggregation.

I'd suggest a few things to make it more readable (and maintainable)

if __name__ block

Put script execution content inside the if __name__ == "__main__" block. Read more about why on stack overflow.

Variable naming

While you follow the PEP-8 convention on variable naming, the names still could use a rework, for eg. session instead of just s.

URL or path

URL refers to "Uniform Resource Locator", which is of the form:

scheme:[//authority]path[?query][#fragment]

You are dealing with only the path here, scheme and authority sections have been fixed as https://example.com/. This is again naming convenience.

Gather vs create

You are creating as well as gathering tasks in create_tasks function.

Type hinting

New in python-3.x is the type hinting feature. I suggest using it whenever possible.

Rewrite

import asyncio
import aiohttp

HOST = "https://example.com"
THINGS_FILE = "things.txt"


def validate_path(path: str) -> bool:
    return 5 <= len(path) < 16 and path.isalnum()


async def check_404(session: aiohttp.ClientSession, path: str):
    async with session.head(f"{HOST}/{path}") as response:
        if response.status == 404:
            print("hello i am working")


async def execute_requests(paths: list[str]):
    async with aiohttp.ClientSession() as session:
        tasks = []
        for path in paths:
            if validate_path(path):
                task = asyncio.create_task(check_404(session, path))
                tasks.append(task)
        return await asyncio.gather(*tasks)


def main():
    with open(THINGS_FILE) as things:
        paths = [line.strip().lower() for line in things]
    while True:
        asyncio.get_event_loop().run_until_complete(execute_requests(paths))


if __name__ == "__main__":
    main()

This can be further rewritten depending on the data being read from file, using a map/filter to only iterate over validated paths in file etc. The above is mostly a suggestion.