Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task. While typically task-agnostic in architecture, this method still requires task-specific fine-tuning datasets of thousands or tens of thousands of examples. By contrast, humans can generally perform a new language task from only a few examples or from simple instructions -something which current NLP systems still largely struggle to do. Here we show that scaling up language models greatly improves task-agnostic, few-shot performance, sometimes even reaching competitiveness with prior state-of-the-art finetuning approaches. Specifically, we train GPT-3, an autoregressive language model with 175 billion parameters, 10x more than any previous non-sparse language model, and test its performance in the few-shot setting. For all tasks, GPT-3 is applied without any gradient updates or fine-tuning, with tasks and few-shot demonstrations specified purely via text interaction with the model. GPT-3 achieves strong performance on many NLP datasets, including translation, question-answering, and cloze tasks, as well as several tasks that require on-the-fly reasoning or domain adaptation, such as unscrambling words, using a novel word in a sentence, or performing 3-digit arithmetic. At the same time, we also identify some datasets where GPT-3's few-shot learning still struggles, as well as some datasets where GPT-3 faces methodological issues related to training on large web corpora. Finally, we find that GPT-3 can generate samples of news articles which human evaluators have difficulty distinguishing from articles written by humans. We discuss broader societal impacts of this finding and of GPT-3 in general.
Two-dimensional-layered heterojunctions have attracted extensive interest recently due to their exciting behaviours in electronic/optoelectronic devices as well as solar energy conversion systems. However, layered heterojunction materials, especially those made by stacking different monolayers together by strong chemical bonds rather than by weak van der Waal interactions, are still challenging to fabricate. Here the monolayer Bi2WO6 with a sandwich substructure of [BiO]+–[WO4]2−–[BiO]+ is reported. This material may be characterized as a layered heterojunction with different monolayer oxides held together by chemical bonds. Coordinatively unsaturated Bi atoms are present as active sites on the surface. On irradiation, holes are generated directly on the active surface layer and electrons in the middle layer, which leads to the outstanding performances of the monolayer material in solar energy conversion. Our work provides a general bottom-up route for designing and preparing novel monolayer materials with ultrafast charge separation and active surface.
The mechanism of photocatalytic conversion of CO(2) and H(2)O over copper oxide promoted titania, Cu(I)/TiO(2), was investigated by means of in situ DRIFT spectroscopy in combination with isotopically labeled (13)CO(2). In addition to small amounts of (13)CO, (12)CO was demonstrated to be the primary product of the reaction by the 2115 cm(-1) Cu(I)-CO signature, indicating that carbon residues on the catalyst surface are involved in reactions with predominantly photocatalytically activated surface adsorbed water. This was confirmed by prolonged exposure of the catalyst to light and water vapor, which significantly reduced the amount of CO formed in a subsequent experiment in the DRIFT cell. In addition, formation of carboxylates and (bi)carbonates was observed by exposure of the Cu(I)/TiO(2) surface to CO(2) in the dark. These carboxylates and (bi)carbonates decompose upon light irradiation, yielding predominantly CO(2). At the same time a novel carbonate species is produced (having a main absorption at approximately 1395 cm(-1)) by adsorption of photocatalytically produced CO on the Cu(I)/TiO(2) surface, most likely through a reverse Boudouard reaction of photocatalytically activated CO(2) with carbon residues. The finding that carbon residues are involved in photocatalytic water activation and CO(2) reduction might have important implications for the rates of artificial photosynthesis reported in many studies in the literature, in particular those using photoactive materials synthesized with carbon containing precursors.
A marigold-like SiC@MoS 2 nanoflower with a unique Z-scheme structure efficiently achieves the overall conversion of gas phase CO 2 with H 2 O (CO 2 (g) + 2H 2 O (g) = CH 4 + 2O 2 ) without any sacrificial reagents under visible light (λ ≥ 420 nm) irradiation. The CH 4 and O 2 evolution are 323 and 621 μL•g −1 •h −1 , and stable throughout 5 cycle reactions of total 40 h. This work demonstrates a breakthrough in artificial photosynthesis with the Z-scheme 1D heterojunction constructed by combining 2D semiconductor and 3D semiconductor based on the transfer balance of photogenerated electron and hole.
The effect of surface plasmon resonance (SPR) on the photocatalytic water splitting was studied by employing the photocatalyst, Au/TiO2, to produce renewable solar hydrogen. It is well-known that metal particles on TiO2 can behave as electron traps, retarding the recombination of electron−hole pairs, thereby improving reaction activity. However, the electron trap is not the only mechanism responsible for the photoreaction enhancement. Our experiment on methylene blue photodegradation over Au particles proved that the SPR phenomenon was also involved in the photoreaction enhancement. Furthermore, the photocatalytic water splitting was performed on Au/TiO2 prepared by the photodeposition method. The production of hydrogen was significantly increased because Au particles not only acted as electron traps as well as active sites but also played an important role in the SPR enhancement. The intensified electric field at the interface between the Au particle and the subdomain on TiO2 was illustrated by finite element method (FEM) electromagnetic simulation.
The stoichiometric photocatalytic reaction of CO 2 with H 2 O is one of the great challenges in photocatalysis. Here, we construct a Cu 2 O-Pt/SiC/IrO x composite by a controlled photodeposition and then an artificial photosynthetic system with Nafion membrane as diaphragm separating reduction and oxidation half-reactions. The artificial system exhibits excellent photocatalytic performance for CO 2 reduction to HCOOH and H 2 O oxidation to O 2 under visible light irradiation. The yields of HCOOH and O 2 meet almost stoichiometric ratio and are as high as 896.7 and 440.7 μmol g −1 h −1 , respectively. The high efficiencies of CO 2 reduction and H 2 O oxidation in the artificial system are attributed to both the direct Z-scheme electronic structure of Cu 2 O-Pt/SiC/IrO x and the indirect Z-scheme spatially separated reduction and oxidation units, which greatly prolong lifetime of photogenerated electrons and holes and prevent the backward reaction of products. This work provides an effective and feasible strategy to increase the efficiency of artificial photosynthesis.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.