LM Arena is a particularly bad comparison site too. Prompts that they use are usually incredibly generic like "A digital render of a sleek, futuristic motorcycl

LM Arena is a particularly bad comparison site too. Prompts that they use are usually incredibly generic like "A digital render of a sleek, futuristic motorcycle racing through a neon-lit cityscape." I actually built GenAI Showdown a while back because I was deeply unsatisfied with LM Arena and other purported comparison tables which either (A) relied solely on visual fidelity (which is a far less interesting benchmark than adherence, IMHO) and/or (B) relied on extremely simplistic and banal prompts.