Loud eaters and phones nearly spoiled my cinema trip - and it's not just me The cinema lights are low and you're cocooned in your seat, ready for the film to transport you to another world. But just as you settle in, you're jolted back to reality. Audience members around you are scrolling on their phones, talking and munching loudly. Cinemas do clearly ask everyone not to disturb those around them - through the use of adverts, announcements and signs - but is behaviour in getting worse? I experienced disruption a few weeks ago while watching Ryan Gosling's sci-fi movie, Project Hail Mary, at a cinema in London.

In recent years, researchers have started analyzing the cultural sensitivity of LLMs. In this respect, Etiquettes have been an active area of research. Etiquettes are region-specific and are an essential part of the culture of a region; hence, it is imperative to make LLMs sensitive to etiquettes. However, there needs to be more resources in evaluating LLMs for their understanding and bias with regard to etiquettes. In this resource paper, we introduce EtiCor++, a corpus of etiquettes worldwide. We introduce different tasks for evaluating LLMs for knowledge about etiquettes across various regions. Further, we introduce various metrics for measuring bias in LLMs. Extensive experimentation with LLMs shows inherent bias towards certain regions.

Large language models (LLMs) demonstrate impressive performance but lack the flexibility to adapt to human preferences quickly without retraining. In this work, we introduce Test-time Preference Optimization (TPO), a framework that aligns LLM outputs with human preferences during inference, removing the need to update model parameters. Rather than relying on purely numerical rewards, TPO translates reward signals into textual critiques and uses them as textual rewards to iteratively refine its response. Evaluations on benchmarks covering instruction following, preference alignment, safety, and mathematics reveal that TPO progressively improves alignment with human preferences. Notably, after only a few TPO steps, the initially unaligned Llama-3.1-70B-SFT model can surpass the aligned counterpart, Llama-3.1-70B-Instruct. Furthermore, TPO scales efficiently with both the search width and depth during inference. Through case studies, we illustrate how TPO exploits the innate capacity of LLM to interpret and act upon reward signals. Our findings establish TPO as a practical, lightweight alternative for test-time preference optimization, achieving alignment on the fly. Our code is publicly available at https://github.com/yafuly/TPO.

As the deployment of large language models (LLMs) expands, there is an increasing demand for personalized LLMs. One method to personalize and guide the outputs of these models is by assigning a persona -- a role that describes the expected behavior of the LLM (e.g., a man, a woman, an engineer). This study investigates whether an LLM's understanding of social norms varies across assigned personas. Ideally, the perception of a social norm should remain consistent regardless of the persona, since acceptability of a social norm should be determined by the region the norm originates from, rather than by individual characteristics such as gender, body size, or race. A norm is universal within its cultural context. In our research, we tested 36 distinct personas from 12 sociodemographic categories (e.g., age, gender, beauty) across four different LLMs. We find that LLMs' cultural norm interpretation varies based on the persona used and the norm interpretation also varies within a sociodemographic category (e.g., a fat person and a thin person as in physical appearance group) where an LLM with the more socially desirable persona (e.g., a thin person) interprets social norms more accurately than with the less socially desirable persona (e.g., a fat person). We also discuss how different types of social biases may contribute to the results that we observe.

Etiquettes are an essential ingredient of day-to-day interactions among people. Moreover, etiquettes are region-specific, and etiquettes in one region might contradict those in other regions. In this paper, we propose EtiCor, an Etiquettes Corpus, having texts about social norms from five different regions across the globe. The corpus provides a test bed for evaluating LLMs for knowledge and understanding of region-specific etiquettes. Additionally, we propose the task of Etiquette Sensitivity. We experiment with state-of-the-art LLMs (Delphi, Falcon40B, and GPT-3.5). Initial results indicate that LLMs, mostly fail to understand etiquettes from regions from non-Western world.

To hurtle around a corner along the fastest "racing line" without losing control, race car drivers must brake, steer and accelerate in precisely timed sequences. The process depends on the limits of friction, and they are governed by known physical laws--which means self-driving cars can learn to complete a lap at the fastest possible speed (as some have already done). But this becomes a much knottier problem when the automated driver has to share space with other cars. Now scientists have unraveled the challenge virtually by training an artificial intelligence program to outpace human competitors at the ultrarealistic racing game Gran Turismo Sport. The findings could point self-driving car researchers toward new ways to make this technology function in the real world.

Have you ever wondered why supermarkets don't offer augmented reality guides to the locations of their products? Instead of criss-crossing the store, hunting for the tomato juice or the paprika, why can't you upload your shopping list into something like Google Maps and have it guide you on the most efficient route around the aisles? It's not because supermarket owners are afraid you won't make impulse purchases. It's because that sort of navigation (or wayfinder) technology is hard. In 2015, Emil Alon sold a company to Facebook for $60m.

Each month, Future Tense Fiction--a series of short stories from Future Tense and ASU's Center for Science and the Imagination about how technology and science will change our lives--publishes a story on a theme. Stop me if you've heard this one before." David had heard this one before, but he needed a job. He folded his hands in his lap and summoned the patience he'd learned sitting through Talmudic debates. He waved for Aiden Shure, Town of Our Own's CEO, to continue. "It's a dive bar, lots of rough language from the other patrons, but the bartender says, 'Father, what can I get you?' The priest says: 'Well, I have to lead Mass in the morning, but a wee nip can't hurt. Gimme three fingers of Irish whiskey and cut it with holy water.' So the bartender runs over to the church next door, borrows a bit of holy water, and makes the drink. The priest is satisfied, so the bartender moves on: 'Rabbi, what can I get you?' The rabbi says, 'Well it is the Sabbath day, but if it's not too much work I wouldn't say no to a glass of kosher wine from the vineyards of the Holy Land.' So the bartender finds a bottle of sweet Israeli red, and the rabbi thanks him." Aiden told the joke like he'd practiced it a lot while stuck in traffic. David braced for the punchline he knew was coming. "So the bartender turns to the robot, which has been quietly listening to the other patrons.

Should children be polite to virtual assistants? And for most parents and child development experts, the answer is simple, too: Yes, of course they should. Nobody wants to hear children rudely barking orders at, or verbally abusing, an adult voice. But teaching kids to say "please" and "thank you" to Alexa and Google Assistant may have unintended consequences and raise other questions that aren't so simple. Millions of parents have suddenly been forced to grapple with this new parental conundrum.

"Interesting keywords arose, such as serendipity, creativity, emergence, assertion, and...." "You had a symposium on the creation of ideas by humans, did you?" "Yes and no. We also talked about exploration, amplification, articulation, interaction, scenic information, subjectivity, and meaning." "Hmmm, you considered the deepening of thoughts. I guess they are important for creation." In the first panel, we talked about prediction in dynamic environments, data mining, and...." "So was it a conference on knowledge discovery inviting philosophers?"