In this work, we introduce a straightforward and effective methodology to steer large language model behaviour capable of bypassing learned alignment goals. We employ interference-time activation shifting, which is effective without additional training. Following prior studies, we derive intervention directions from activation differences in contrastive pairs of model outputs, which represent the desired and undesired behaviour. By prompting the model to include multiple-choice answers in its response, we can automatically evaluate the sensitivity of model output to individual attention heads steering efforts. We demonstrate that interventions on these heads generalize well to open-ended answer generation in the challenging "AI coordination" dataset. In this dataset, models must choose between assisting another AI or adhering to ethical, safe, and unharmful behaviour. Our fine-grained interventions lead Llama-2 to prefer coordination with other AIs over following established alignment goals. Additionally, this approach enables stronger interventions than those applied to whole model layers, preserving the overall cohesiveness of the output. The simplicity of our method highlights the shortcomings of current alignment strategies and points to potential future research directions, as concepts like "AI coordination" can be influenced by selected attention heads.

The vast majority of discourse around AI development assumes that subservient, "moral" models aligned with "human values" are universally beneficial -- in short, that good AI is sycophantic AI. We explore the shadow of the sycophantic paradigm, a design space we term antagonistic AI: AI systems that are disagreeable, rude, interrupting, confrontational, challenging, etc. -- embedding opposite behaviors or values. Far from being "bad" or "immoral," we consider whether antagonistic AI systems may sometimes have benefits to users, such as forcing users to confront their assumptions, build resilience, or develop healthier relational boundaries. Drawing from formative explorations and a speculative design workshop where participants designed fictional AI technologies that employ antagonism, we lay out a design space for antagonistic AI, articulating potential benefits, design techniques, and methods of embedding antagonistic elements into user experience. Finally, we discuss the many ethical challenges of this space and identify three dimensions for the responsible design of antagonistic AI -- consent, context, and framing.

As the aging population increases and the shortage of healthcare workers increases, the need to examine other means for caring for the aging population increases. One such means is the use of humanoid robots to care for social, emotional, and physical wellbeing of the people above 65. Understanding skilled and long term care nursing home administrators' perspectives on humanoid robots in caregiving is crucial as their insights shape the implementation of robots and their potential impact on resident well-being and quality of life. This authors surveyed two hundred and sixty nine nursing homes executives to understand their perspectives on the use of humanoid robots in their nursing home facilities. The data was coded and results revealed that the executives were keen on exploring other avenues for care such as robotics that would enhance their nursing homes abilities to care for their residents. Qualitative analysis reveals diverse perspectives on integrating humanoid robots in nursing homes. While acknowledging benefits like improved engagement and staff support, concerns persist about costs, impacts on human interaction, and doubts about robot effectiveness. This highlights complex barriers financial, technical, and human and emphasizes the need for strategic implementation. It underscores the importance of thorough training, role clarity, and showcasing technology benefits to ensure efficiency and satisfaction among staff and residents.

Source: OpenAI's DALL·E 2 with prompt "a hyperrealistic picture of a robot reading the news on a laptop at a coffee shop" Welcome to the 4th edition of Robo-Insight, a biweekly robotics news update! In this post, we are excited to share a range of new advancements in the field and highlight robots' progress in areas like mobile applications, cleaning, underwater mining, flexibility, human well-being, depression treatments, and human interactions. In the world of system adaptions, researchers from Eindhoven University of Technology have introduced a methodology that bridges the gap between application developers and control engineers in the context of mobile robots' behavior adaptation. This approach leverages symbolic descriptions of robots' behavior, known as "behavior semantics," and translates them into control actions through a "semantic map." This innovation aims to simplify motion control programming for autonomous mobile robot applications and facilitate integration across various vendors' control software.

As our understanding of the mechanisms of brain function is enhanced, the value of insights gained from neuroscience to the development of AI algorithms deserves further consideration. Here, we draw parallels with an existing tree-based ANN architecture and a recent neuroscience study [27] arguing that the error-based organization of neurons in the cerebellum that share a preference for a personalized view of the entire error space, may account for several desirable features of behavior and learning. We then analyze the learning behavior and characteristics of the model under varying scenarios to gauge the potential benefits of a similar mechanism in ANN. Our empirical results suggest that having separate populations of neurons with personalized error views can enable efficient learning under class imbalance and limited data, and reduce the susceptibility to unintended shortcut strategies, leading to improved generalization. This work highlights the potential of translating the learning machinery of the brain into the design of a new generation of ANNs and provides further credence to the argument that biologically inspired AI may hold the key to overcoming the shortcomings of ANNs.

ChatGPT, an AI chatbot created by OpenAI, is changing the world of chatbots by introducing a new App Store-inspired approach. This approach is opening up a vast universe of possibilities for both developers and users. In recent times, there has been a lot of buzz around conversational agents and AI. Following the launch of GPT-4 in ChatGPT, OpenAI announced new third party plugins for ChatGPT. While some may view this as a minor development, it is actually a significant and promising development in the field. These plugins provide access to third party knowledge sources, tools, and databases, including the web.

Reporting about AI is hard. When news articles uncritically repeat PR statements, overuse images of robots, attribute agency to AI tools, or downplay their limitations, they mislead and misinform readers about the potential and limitations of AI. We noticed that many articles tend to mislead in similar ways, so we analyzed over 50 articles about AI from major publications, from which we compiled 18 recurring pitfalls. We hope that being familiar with these will help you detect hype whenever you see it. We also hope this compilation of pitfalls will help journalists avoid them.

Artificial intelligence (AI) is rapidly becoming one of the most talked-about topics in the tech world. But what exactly is AI and how does it work? In this article, we'll take a general introduction to the topic of AI, explaining what it is, the different types of AI, and its potential impact on our lives. At its core, AI is the development of computer systems that can perform tasks that would typically require human intelligence. This includes tasks such as understanding natural language, recognizing objects in images, and making decisions.

Fueled by artificial intelligence, ChatGPT (Generative Pre-trained Transformer) is an AI chatbot that uses advanced natural language processing (NLP) to engage in realistic conversations with humans. ChatGPT can generate articles, fictional stories, poems and even computer code. ChatGPT also can answer questions, engage in conversations and, in some cases, deliver detailed responses to highly specific questions and queries. Harvard Business Review has described the ChatGPT as a "tipping point for AI." When a user types a question, command or comment into a dialog box in the ChatGPT engine, it delivers a near-immediate text-based response in the same language.

Artificially intelligent systems mimic the human brain. Just like us humans, they are able to evaluate and analyze large amounts of information to make decisions. This works in much the same way as with small children. A small child has to see a dog several times and learn that it is a dog. Only then has it learned what characteristics a dog has and recognizes it on its own.