This study explores human-robot interaction (HRI) based on a mobile robot and YOLO to increase real-time situation awareness and prevent accidents in the workplace. Using object segmentation, we propose an approach that is capable of analyzing these situations in real-time and providing useful information to avoid critical working situations. In the industry, ensuring the safety of workers is paramount, and solutions based on robots and AI can provide a safer environment. For that, we proposed a methodology evaluated with two different YOLO versions (YOLOv8 and YOLOv5) alongside a LoCoBot robot for supervision and to perform the interaction with a user. We show that our proposed approach is capable of navigating a test scenario and issuing alerts via Text-to-Speech when dangerous situations are faced, such as when hardhats and safety vests are not detected. Based on the results gathered, we can conclude that our system is capable of detecting and informing risk situations such as helmet/no helmet and safety vest/no safety vest situations.

Recently, there has been a substantial amount of interest in GNN-based anomaly detection. Existing efforts have focused on simultaneously mastering the node representations and the classifier necessary for identifying abnormalities with relatively shallow models to create an embedding. Therefore, the existing state-of-the-art models are incapable of capturing nonlinear network information and producing suboptimal outcomes. In this thesis, we deploy decoupled GNNs to overcome this issue. Specifically, we decouple the essential node representations and classifier for detecting anomalies. In addition, for node representation learning, we develop a GNN architecture with two modules for aggregating node feature information to produce the final node embedding. Finally, we conduct empirical experiments to verify the effectiveness of our proposed approach. The findings demonstrate that decoupled training along with the global context enhanced representation of the nodes is superior to the state-of-the-art models in terms of AUC and introduces a novel way of capturing the node information.

Two primary ways to change LLM behavior are prompting and weight updates (e.g., fine-tuning). Prompting LLMs is simple and effective, specifying the desired changes explicitly in natural language, whereas weight updates provide more expressive and permanent behavior changes, specified implicitly via training on large datasets. We present a technique for "baking" prompts into the weights of an LLM. Prompt Baking converts a prompt $u$ and initial weights $\theta$ to a new set of weights $\theta_u$ such that new "baked" LLM behaves like the original prompted LLM. Mathematically, we minimize the KL divergence between $P_\theta(\cdot | u)$ and $P_{\theta_u}(\cdot)$, where $P$ is the LLM's probability distribution over token sequences. Across all our experiments, we find prompts can be readily baked into weight updates. Baking chain-of-thought prompts improves zero-shot performance on GSM8K, ASDiv, MBPP, ARC-Easy, ARC-Challenge, and CommonsenseQA benchmarks. Baking news headlines directly updates an LLM's knowledge. And baking instructions & personas alleviates "prompt forgetting" over long sequences. Furthermore, stopping baking early creates "half-baked" models, continuously scaling prompt strength. Baked models retain their sensitivity to further prompting and baking, including re-prompting with the baked-in prompt. Surprisingly, the re-prompted models yield further performance gains in instruction following, as well as math reasoning and coding benchmarks. Taking re-prompting and re-baking to the limit yields a form of iterative self-improvement we call Prompt Pursuit, and preliminary results on instruction following exhibit dramatic performance gains. Finally, we discuss implications for AI safety, continuous model updating, enhancing real-time learning capabilities in LLM-based agents, and generating more stable AI personas.

While there will be more elections this year where AI could have an effect, the United States being one likely to attract particular attention, the trend observed thus far is unlikely to change. AI is being used to try to influence electoral processes, but these efforts have not been fruitful. Commenting on the upcoming US election, Meta's latest Adversarial Threat Report acknowledged that AI was being used to meddle--for example, by Russia-based operations--but that "GenAI-powered tactics provide only incremental productivity and content-generation gains" to such "threat actors." This echoes comments from the company's president of global affairs, Nick Clegg, who earlier this year stated that "it is striking how little these tools have been used on a systematic basis to really try to subvert and disrupt the elections." Far from being dominated by AI-enabled catastrophes, this election "super year" at that point was pretty much like every other election year.

The South Korean police began investigating messaging app Telegram for the first time for its alleged negligence over deepfake-related sex crimes, Yonhap News reported, citing a briefing from the head of National Office of Investigation. The probe was launched by the Seoul Metropolitan Police and like in other countries, Telegram has been uncooperative in sharing materials with the authorities, investigation office head Woo Jong-soo said, according to Yonhap on Monday. South Korea is doing its best to deploy its own investigating tools to extract the materials, Yonhap said, citing Woo. The investigation comes less than a week after the South Korean government asked Telegram and other social media companies to delete deepfake images from their platforms after an increase in such content. President Yoon Suk-yeol last week described deepfake as a "clear crime" and ordered officials to take stern actions to counter it.

Among the many candidates for the ruling Liberal Democratic Party's leadership race, which effectively decides who Japan's next prime minister will be, an anti-nuclear policy stance is fading. Some candidates who were once known as anti-nuclear advocates are shifting their position as energy demands are likely to see a big jump due to developments in artificial intelligence and the need to build more data centers to power such technology. "As for energy, it would be a huge blow to the economy if companies choose to move their data centers outside of Japan and invest in AI in other countries, just because Japan is unable to provide electricity," said digital minister Taro Kono, who recently declared his candidacy for the Sept. 27 LDP leadership race.

Recent large language model (LLM) defenses have greatly improved models' ability to refuse harmful queries, even when adversarially attacked. However, LLM defenses are primarily evaluated against automated adversarial attacks in a single turn of conversation, an insufficient threat model for real-world malicious use. We demonstrate that multi-turn human jailbreaks uncover significant vulnerabilities, exceeding 70% attack success rate (ASR) on HarmBench against defenses that report single-digit ASRs with automated single-turn attacks. Human jailbreaks also reveal vulnerabilities in machine unlearning defenses, successfully recovering dual-use biosecurity knowledge from unlearned models. We compile these results into Multi-Turn Human Jailbreaks (MHJ), a dataset of 2,912 prompts across 537 multi-turn jailbreaks. We publicly release MHJ alongside a compendium of jailbreak tactics developed across dozens of commercial red teaming engagements, supporting research towards stronger LLM defenses.

The Game of Life (Life), a well known algorithm within the broader class of cellular automata (CA), exhibits complex emergent dynamics, with extreme sensitivity to initial conditions. Modeling and predicting such intricate behavior without explicit knowledge of the system's underlying topology presents a significant challenge, motivating the development of algorithms that can generalize across various grid configurations and boundary conditions. We develop a decoder-only generative pretrained transformer model to solve this problem, showing that our model can simulate Life on a toroidal grid with no prior knowledge on the size of the grid, or its periodic boundary conditions (LifeGPT). LifeGPT is topology-agnostic with respect to its training data and our results show that a GPT model is capable of capturing the deterministic rules of a Turing-complete system with near-perfect accuracy, given sufficiently diverse training data. We also introduce the idea of an'autoregressive autoregressor' to recursively implement Life using LifeGPT. Our results pave the path towards true universal computation within a large language model (LLM) framework, synthesizing of mathematical analysis with natural language processing, and probing AI systems for situational awareness about the evolution of such algorithms without ever having to compute them. Similar GPTs could potentially solve inverse problems in multicellular self-assembly by extracting CA-compatible rulesets from realworld biological systems to create new predictive models, which would have significant consequences for the fields of bioinspired materials, tissue engineering, and architected materials design. Cellular automata (CA) have long been a subject of profound interest within the fields of computer science and mathematics, owing to their intricate and emergent behaviors. CA algorithms are uniquely characterized by their combination of computational simplicity--evolving solely by local state-transition rules--and broad dynamical behavior, encompassing static, periodic, chaotic, and complex patterns, depending on the ruleset and initial condition (IC) being used. These properties render CA algorithms particularly valuable for simulating a wide array of natural phenomena, such as the propagation of forest fires[1], traffic flow dynamics[2], chemical reactions[3], and recrystallization[4]. The inherent behavioral unpredictability of CA (following human inspection of their rulesets) has hindered advancements in these subfields, confining CA to the realm of phenomenological modeling and subsequently preventing their evolution into mature, predictive tools for systems or phenomena where do not yet know a closed-form ruleset.

Malicious shell commands are linchpins to many cyber-attacks, but may not be easy to understand by security analysts due to complicated and often disguised code structures. Advances in large language models (LLMs) have unlocked the possibility of generating understandable explanations for shell commands. However, existing general-purpose LLMs suffer from a lack of expert knowledge and a tendency to hallucinate in the task of shell command explanation. In this paper, we present Raconteur, a knowledgeable, expressive and portable shell command explainer powered by LLM. Raconteur is infused with professional knowledge to provide comprehensive explanations on shell commands, including not only what the command does (i.e., behavior) but also why the command does it (i.e., purpose). To shed light on the high-level intent of the command, we also translate the natural-language-based explanation into standard technique & tactic defined by MITRE ATT&CK, the worldwide knowledge base of cybersecurity. To enable Raconteur to explain unseen private commands, we further develop a documentation retriever to obtain relevant information from complementary documentations to assist the explanation process. We have created a large-scale dataset for training and conducted extensive experiments to evaluate the capability of Raconteur in shell command explanation. The experiments verify that Raconteur is able to provide high-quality explanations and in-depth insight of the intent of the command.

Social scientists quickly adopted large language models due to their ability to annotate documents without supervised training, an ability known as zero-shot learning. However, due to their compute demands, cost, and often proprietary nature, these models are often at odds with replication and open science standards. This paper introduces the Political DEBATE (DeBERTa Algorithm for Textual Entailment) language models for zero-shot and few-shot classification of political documents. These models are not only as good, or better than, state-of-the art large language models at zero and few-shot classification, but are orders of magnitude more efficient and completely open source. By training the models on a simple random sample of 10-25 documents, they can outperform supervised classifiers trained on hundreds or thousands of documents and state-of-the-art generative models with complex, engineered prompts. Additionally, we release the PolNLI dataset used to train these models -- a corpus of over 200,000 political documents with highly accurate labels across over 800 classification tasks.