You can actually talk to the candidate about their solution: why they picked one approach over another, why X wouldn't work, how well they use LLMs (e.g. 10-line comments for a single line of simple code? do they reuse functions or create four different clones?), etc. With LLMs, assignments can be larger/more complex so you can see more.

I'm not saying it's better or worse, but I don't think assignments are inherently bad. It works the same way in education from what I've heard - written assignments may not be very valuable on their own anymore, but you can ask students questions about specific things irl to see if they actually know their stuff.