Frontier AI systems, including large language models (LLMs), hold increasing influence over the epistemology of human users. Such influence can reinforce prevailing societal values, potentially contributing to the lock-in of misguided moral beliefs and, consequently, the perpetuation of problematic moral practices on a broad scale. We introduce progress alignment as a technical solution to mitigate this imminent risk. Progress alignment algorithms learn to emulate the mechanics of human moral progress, thereby addressing the susceptibility of existing alignment methods to contemporary moral blindspots.

Frontier AI systems, including large language models (LLMs), hold increasing influence over the epistemology of human users. Such influence can reinforce prevailing societal values, potentially contributing to the lock-in of misguided moral beliefs and, consequently, the perpetuation of problematic moral practices on a broad scale. We introduce progress alignment as a technical solution to mitigate this imminent risk. Progress alignment algorithms learn to emulate the mechanics of human moral progress, thereby addressing the susceptibility of existing alignment methods to contemporary moral blindspots. To empower research in progress alignment, we introduce ProgressGym, an experimental framework allowing the learning of moral progress mechanics from history, in order to facilitate future progress in real-world moral decisions. Leveraging 9 centuries of historical text and 18 historical LLMs, ProgressGym enables codification of real-world progress alignment challenges into concrete benchmarks. Specifically, we introduce three core challenges: tracking evolving values (PG-Follow), preemptively anticipating moral progress (PG-Predict), and regulating the feedback loop between human and AI value shifts (PG-Coevolve). Alignment methods without a temporal dimension are inapplicable to these tasks. In response, we present lifelong and extrapolative algorithms as baseline methods of progress alignment, and build an open leaderboard soliciting novel algorithms and challenges. The framework and the leaderboard are available at https://github.com/PKU-Alignment/ProgressGym and https://huggingface.co/spaces/PKU-Alignment/ProgressGym-LeaderBoard respectively.

News captioning task aims to generate sentences by describing named entities or concrete events for an image with its news article. Existing methods have achieved remarkable results by relying on the large-scale pre-trained models, which primarily focus on the correlations between the input news content and the output predictions. However, the news captioning requires adhering to some fundamental rules of news reporting, such as accurately describing the individuals and actions associated with the event. In this paper, we propose the rule-driven news captioning method, which can generate image descriptions following designated rule signal. Specifically, we first design the news-aware semantic rule for the descriptions. This rule incorporates the primary action depicted in the image (e.g., "performing") and the roles played by named entities involved in the action (e.g., "Agent" and "Place"). Second, we inject this semantic rule into the large-scale pre-trained model, BART, with the prefix-tuning strategy, where multiple encoder layers are embedded with news-aware semantic rule. Finally, we can effectively guide BART to generate news sentences that comply with the designated rule. Extensive experiments on two widely used datasets (i.e., GoodNews and NYTimes800k) demonstrate the effectiveness of our method.

One evening in September of 1957, viewers across America could turn on their television sets and tune in to a CBS broadcast during which a young woman dropped acid. She sat next to a man in a suit: Sidney Cohen, the researcher who had given her the LSD. The woman wore lipstick and nail polish, and her eyes were shining. "I wish I could talk in Technicolor," she said. And, at another point, "I can see the molecules. Were some families maybe--oh, I don't know--eating meat loaf on TV trays as they watched this nice lady undergo her mind-bending, molecule-revealing journey through inner space? Did they switch to "Father Knows Best" or "The Perry Como Show" afterward? One of the feats that the historian Benjamin Breen pulls off in his lively and engrossing new book, "Tripping on Utopia: Margaret Mead, the Cold War, and the Troubled Birth of Psychedelic Science" (Grand Central), is to make a cultural moment like the anonymous woman's televised trip seem less incongruous, if no less ...

The centuries-old process of releasing defendants on bail, long the province of judicial discretion, is getting a major assist ... courtesy of artificial intelligence. In late August, Hercules Shepherd Jr. walked up to the stand in a Cleveland courtroom, dressed in an orange jumpsuit. Two nights earlier, an officer had arrested him at a traffic stop with a small bag of cocaine, and he was about to be arraigned. Judge Jimmy Jackson Jr. looked at Shepherd, then down at a computer-generated score on the front of the 18-year-old's case file. The scores marked Shepherd as a prime candidate for pretrial release with low bail.