As AI capabilities increasingly surpass human proficiency in complex tasks, current alignment techniques including SFT and RLHF face fundamental challenges in ensuring reliable oversight. These methods rely on direct human assessment and become untenable when AI outputs exceed human cognitive thresholds. In response to this challenge, we explore two hypotheses: (1) critique of critique can be easier than critique itself, extending the widely-accepted observation that verification is easier than generation to the critique domain, as critique itself is a specialized form of generation; (2) this difficulty relationship is recursively held, suggesting that when direct evaluation is infeasible, performing high-order critiques (e.g., critique of critique of critique) offers a more tractable supervision pathway. To examine these hypotheses, we perform Human-Human, Human-AI, and AI-AI experiments across multiple tasks. Our results demonstrate encouraging evidence supporting these hypotheses and suggest that recursive self-critiquing is a promising direction for scalable oversight.

Fox News Flash top headlines are here. Check out what's clicking on Foxnews.com. Israel acknowledged on Friday that an inscription in clay found in the country's south bearing the name of Darius the Great, ruler of the ancient Persian Empire, was not authentic. The shard of pottery in question was discovered by a passerby last December and caused a sensation as the first mention of sixth century B.C. empire builder to appear in Israel. After the news broke earlier this week, an expert in ancient Aramaic inscriptions approached the Israel Antiquities Authority to explain that she herself had actually etched those words onto the ancient fragment.

Archeologists at Northern Arizona University (NAU) have taught computers to sort pottery fragments by design and style to assist in classification and reconstruction. A team at NAU's department of anthropology used a form of machine learning known as Convolutional Neural Networks (CNNs) to create a computerized method that emulates the thought processes of the human mind when it analyzes visual information to rapidly and consistently sort thousands of pottery designs into categories. CNNs are commonly used in computer image recognition processes like comparing X-rays to medical conditions, matching images in search engines and in self-driving cars. "Now, using digital photographs of pottery, computers can accomplish what used to involve hundreds of hours of tedious, painstaking and eye-straining work by archaeologists who physically sorted pieces of broken pottery into groups, in a fraction of the time and with greater consistency," said study author Leszek Pawlowicz, in a release. The research results are due to be published in the June edition of the Journal of Archeological Science.

A robot has photographed a nearly intact ancient urn at the bottom of Japan's largest freshwater lake, according to Japanese national paper the Asahi Shimbun. Over the last century, a number of pottery pieces representing a huge range in timeline have been recovered from Lake Biwako, in central Japan. Archaeologists have no idea why. This urn is an example of Haji pottery, earthenware characterized by a rusty reddish-brown color that came from being baked. It measures roughly 12 to 16 inches tall, with the opening at the top measuring roughly 8 inches across, and it likely dates to the seventh or eighth century, according to the newspaper Asahi Shimbun.