The author presents a memorial to various AI benchmarks that have been “killed” by AI advancements over the years. Each benchmark, from reasoning tasks to mathematics problems to common sense evaluations, showcased the progress and limitations of AI models. The website highlights the original scores of the benchmarks and the models that ultimately defeated them. The content reflects on the astonishing advancements made in AI technologies, challenging the notion of whether AI can perform certain tasks. The project acknowledges the difficulty in accurately collecting and attributing data on these benchmarks, and invites contributions to improve accuracy. The website is a tribute to the evolution of AI capabilities and the ongoing quest for machine intelligence.
https://r0bk.github.io/killedbyllm/