Warning: this paper contains model outputs exhibiting offensiveness and biases.Large language models (LLMs) have achieved impressive performance on various natural language generation tasks. Nonetheless, they suffer from generating negative and harmful contents that are biased against certain demographic groups (e.g., female), raising severe fairness concerns. As remedies, prior works intervened the generation by removing attitude or demographic information, inevitably degrading the generation quality and resulting in notable fairness-fluency trade-offs. However, it is still under-explored to what extent the fluency has to be affected in order to achieve a desired level of fairness. In this work, we conduct the first formal study from an information-theoretic perspective. We show that previous approaches are excessive for debiasing and propose LIDAO, a general framework to debias a (L)LM at a better fluency provably. We further robustify LIDAO in adversarial scenarios, where a carefully-crafted prompt may stimulate LLMs exhibiting instruction-following abilities to generate texts with fairness issue appears only when the prompt is also taken into account. Experiments on three LMs ranging from 0.7B to 7B parameters demonstrate the superiority of our method.

No abstract

The commonly-used superstrate configuration (depositing front subcell first and then depositing back subcell) in all-perovskite tandem solar cells is disadvantageous for long-term stability due to oxidizable narrow-bandgap perovskite assembled last and easily exposable to air. Here we reverse the processing order and demonstrate all-perovskite tandems in a substrate configuration (depositing back subcell first and then depositing front subcell) to bury oxidizable narrow-bandgap perovskite deep in the device stack. By using guanidinium tetrafluoroborate additive in wide-bandgap perovskite subcell, we achieve an efficiency of 25.3% for the substrate-configured all-perovskite tandem cells. The unencapsulated devices exhibit no performance degradation after storage in dry air for 1000 hours. The substrate configuration also widens the choice of flexible substrates: we achieve 24.1% and 20.3% efficient flexible all-perovskite tandem solar cells on copper-coated polyethylene naphthalene and copper metal foil, respectively. Substrate configuration offers a promising route to unleash the commercial potential of all-perovskite tandem solar cells.

With the rise of 2D materials, copper (Cu) is revealed as good surface catalyst, especially in the self‐limited growth of graphene. In the regime of surface catalyst, the catalytic activities and functionalities of Cu should be highly dependent on its surface type. Traditional methods to determine the surface index are mainly high‐vacuum based surface science techniques and are typically of low throughput and in small scale. A method to fast detect the surface index of Cu in large scale is still lacking. Here, the authors report an effective optical contrast method to identify the Cu surface index in large area. This method is based on the Cu2O‐thickness dependent color of Cu surface after a mild oxidation in hot air. It is revealed that different Cu surfaces (Cu(111), Cu(100), and Cu(110) as examples) have various oxidation barriers and would exhibit distinct color evolution with heating time. It is also showed that graphene grown on Cu surfaces with varied orientations has totally different growth behaviors. The results would greatly facilitate the high‐throughput determination of Cu surface index and accelerate the large‐scale facet‐dependent catalytic research of Cu, such as in single‐crystal graphene growth.

Carrier dynamics, the most fundamental process in electronics and optoelectronics, has drawn great attentions owing to its crucial role in property engineering of materials. Exploration and regulation of carrier dynamics are essential for designing devices with specific functions and optimizing their performances. However, the lack of conventional tools with simultaneous ultrafast temporal and ultrasmall spatial resolution has impeded direct observation and manipulation of carrier dynamics at both the femtosecond and nanometer scale. In this study, the direct observation and modulation of ultrafast carrier dynamics at the graphene/gallium arsenide (GaAs) interface is achieved by tuning the doping level of bulk GaAs. This successful characterization is performed using advanced in situ photoemission electron microscopy combined with the ultrafast pump–probe technique. It is found that a change in the doping level in GaAs can change its band bending and switch the hot‐carrier transfer direction at the graphene/GaAs interface with a lifetime reduction of nearly six times. This work paves the way of engineering ultrafast carrier dynamics at 2D interfaces by modifying the 3D bulk properties, and also provides a platform for fundamental studies of ultrafast physics with high spatial resolution.