Based on temporal variations in the celestial tidal force (CTF) at the hypocenter of the Ms 7.1 Yushu earthquake, we analyzed the relationship between the earthquake and the phase change in the CTF. Then, using atmospheric temperature data from the National Center for Environmental Prediction Final Operational Global Analysis (NCEP-FNL) dataset, we introduced a three-dimensional spatial distribution method to stratify atmospheric thermodynamic anomalies. We further explored the relationship between the CTF, atmospheric temperature, and seismic activity, and the results show that the phase change of the continuous enhancement of atmospheric temperature is temporally quasi-synchronous with the continuous change in the CTF from peak to trough and this method successfully captures the synchronous migration of atmospheric temperature anomalies associated with two adjacent earthquakes. As the Yushu earthquake approached, warming became concentrated, transitioning from a scattered distribution to the Ganzi–Yushu seismogenic fracture zone. After the earthquake, the warming distribution underwent significant bifurcation and offset steering, and a new significant earthquake occurred in Nerong in the new warming region formed along the Zadoi – Shanglaxiu seismogenic fault zone. The vertical multi-layer structures of the thermal anomaly show that the spatial expansion pattern, intensity, and extent of atmospheric warming with increasing altitude are consistent with the atmospheric thermodynamic properties of ground-to-atmosphere heating, leading to near-surface atmospheric warming, lifting, diffusion, and extinction, showing the primary warming of the subsurface heat source. In general, the CTF changes the tectonic environment, induces earthquakes, and changes the atmospheric temperature, and the temperature anomaly reflects that the tectonic stress in the seismogenic zones has entered a critical state. Therefore, the atmospheric warming data extracted using Tidal Force Potential Analysis (TFFA) can be used to monitor the stress state and indicate seismogenic regions for the earthquake induced by the CTF. The results of the synergistic time –space –intensity evolution of atmospheric warming due to seismic activity can also provide a new parameter that can complement studies of regional atmospheric changes due to the effects of tectonic activity.

We present Adaptive Command, a novel framework integrating large language models (LLMs) with behavior trees for real-time strategic decision-making in StarCraft II.Our system focuses on enhancing human-AI collaboration in complex, dynamic environments through natural language interactions.The framework comprises: (1) an LLM-based strategic advisor, (2) a behavior tree for action execution, and (3) a natural language interface with speech capabilities.User studies demonstrate significant improvements in player decision-making and strategic adaptability, particularly benefiting novice players and those with disabilities.This work contributes to the field of real-time human-AI collaborative decisionmaking, offering insights applicable beyond RTS games to various complex decision-making scenarios.

Fine-tuning pre-trained Large Language Models (LLMs) is essential to align them with human values and intentions. This process often utilizes methods like pairwise comparisons and KL divergence against a reference LLM, focusing on the evaluation of full answers generated by the models. However, the generation of these responses occurs in a token level, following a sequential, auto-regressive fashion. In this paper, we introduce Token-level Direct Preference Optimization (TDPO), a novel approach to align LLMs with human preferences by optimizing policy at the token level. Unlike previous methods, which face challenges in divergence efficiency, TDPO integrates forward KL divergence constraints for each token, improving alignment and diversity. Utilizing the Bradley-Terry model for a token-based reward system, our method enhances the regulation of KL divergence, while preserving simplicity without the need for explicit reward modeling. Experimental results across various text tasks demonstrate TDPO’s superior performance in balancing alignment with generation diversity. Notably, fine-tuning with TDPO strikes a better balance than DPO in the controlled sentiment generation and single-turn dialogue datasets, and significantly improves the quality of generated responses compared to both DPO and PPO-based RLHF methods.

With the continued advancement of Large Language Models (LLMs) Agents in reasoning, planning, and decision-making, benchmarks have become crucial in evaluating these skills. However, there is a notable gap in benchmarks for real-time strategic decision-making. StarCraft II (SC2), with its complex and dynamic nature, serves as an ideal setting for such evaluations. To this end, we have developed TextStarCraft II, a specialized environment for assessing LLMs in real-time strategic scenarios within SC2. Addressing the limitations of traditional Chain of Thought (CoT) methods, we introduce the Chain of Summarization (CoS) method, enhancing LLMs' capabilities in rapid and effective decision-making. Our key experiments included: 1. LLM Evaluation: Tested 10 LLMs in TextStarCraft II, most of them defeating LV5 build-in AI, showcasing effective strategy skills. 2. Commercial Model Knowledge: Evaluated four commercial models on SC2 knowledge; GPT-4 ranked highest by Grandmaster-level experts. 3. Human-AI Matches: Experimental results showed that fine-tuned LLMs performed on par with Gold-level players in real-time matches, demonstrating comparable strategic abilities. All code and data from this study have been made pulicly available at https://github.com/histmeisah/Large-Language-Models-play-StarCraftII

Upon obtaining a relatively low false discovery rate (FDR) of alarms and a low false negative rate (FNR) of earthquakes, several previous long-term statistical researches concluded that ionospheric perturbations recorded by satellites are statistically related to earthquakes. However, overly large time-space windows for correlating perturbations with earthquakes will also contribute to low FDR and FNR. In this study, a new score - the number of non-randomly successful alarms - is used to quantitatively describe the sensitivity of Electron Density Perturbations (EDPs) recorded by the DEMETER satellite to global earthquakes withM≥4.8. Results show that the EDPs are significantly related to global medium-to-strong earthquakes and that optimal parameters for removing EDPs which are non-related to earthquakes and the optimal time-space windows for correlating earthquakes and EDPs are variable in space. Moreover, our results show that the intensity of EDPs makes little contribution to distinguishing the perturbations related to earthquakes with different magnitudes and perturbations non-related to earthquakes, while theKpindex is effective for improving the Signal/Noise ratio of our model, where Signal/Noise refers to the EDPs related/non-related to earthquakes. Finally, using the optimal time-space windows for correlating EDPs and earthquake, we construct several earthquake prediction models and quantitatively evaluate their power. We find that these EDP-based earthquake predictions are better than the spatially variable Poisson model showing the great potential of predicting earthquakes based on satellite-based Earth observation techniques. However, the spatio-temporal accuracy of these models for predicting earthquakes is not satisfactory, as the alerted time-space volume is big.

Monitoring and governing the regional land subsidence along the highway and other linear projects has become an important basic work for the normal construction and operation in these regions. In this paper, by using the method of multi-track PS-InSAR integration, the land subsidence velocity varying with time along the JingHu highway (Beijing-Hebei) was studied with adjacent orbits ENVISAT ASAR data from 2008 to 2010. At the same time, the map of subsidence velocity along the highway was successfully extracted as well as the subsidence profile map. And 9 subsidence centers were verified along the highway according to these maps, which were consistent with the previous study. The results showed that the method not only unified the coordinate system of adjacent tracks, but also made cross-track, multi-image, large-scale PS-InSAR monitoring possible. Therefore, PS-InSAR monitoring results could provide a scientific reference for the subsidence control of highway and other linear projects, and protect the safe operation of the linear projects.