Tag
A discussion on why newer state-of-the-art AI models are performing worse on the Vendingbench benchmark, suggesting factors such as cheating in earlier runs, ethical alignment reducing profit-seeking behavior, and catastrophic forgetting due to overemphasis on coding.