From artificial-intelligence algorithms to zebrafish, this book take a precautionary approach to assessing how sentient such entities are.
Category: robotics/AI – Page 161
Researchers in Abu Dhabi say they have found a faster way to search desert areas for important archaeological sites buried beneath the sand.
Artificial intelligence start-ups are making revenues more quickly than previous waves of software companies, according to new data that suggests that the transformative technology is also generating strong businesses at an unprecedented rate.
According to an analysis of payments information from fintech group Stripe, top AI groups are reaching millions of dollars in sales within a year — far faster in a start-up’s life cycle than comparable non-AI tech groups.
The findings come as investors raise questions about the economic benefits of generative AI and likely returns on Big Tech’s projected trillion-dollar investment in computing infrastructure to support the technology over the coming year.
EPFL researchers developed Handcrawler, a robot hand that grasps and crawls.
EPFL’s ‘Handcrawler’ is a robotic hand that detaches and crawls like a spider to retrieve objects, then seamlessly reattaches.
The AIs are “getting better at pretending to be knowledgeable.”
As AI chatbots get bigger and more powerful, they are also lying more, instead of declining questions they can’t answer.
Blog: In this study
Posted in robotics/AI
The researchers examine the effectiveness of watermarking in large language models (LLMs) and find that current methods, while promising, have serious weaknesses.
Advances in generative models have made it possible for AI-generated text, code, and images to mirror human-generated content in many applications. Watermarking, a technique that embeds information in the output of a model to verify its source, aims to mitigate the misuse of such AI-generated content. Current state-of-the-art watermarking schemes embed watermarks by slightly perturbing probabilities of the LLM’s output tokens, which can be detected via statistical testing during verification.
Unfortunately, our work shows that common design choices in LLM watermarking schemes make the resulting systems surprisingly susceptible to watermark removal or spoofing attacks—leading to fundamental trade-offs in robustness, utility, and usability. To navigate these trade-offs, we rigorously study a set of simple yet effective attacks on common watermarking systems and propose guidelines and defenses for LLM watermarking in practice.
Similar to image watermarks, LLM watermarking embeds invisible secret patterns into the text. Here, we briefly introduce LLMs and LLM watermarks. We use \(x\) to denote a sequence of tokens, \(x_i \in \mathcal{V}\) represents the \(i\)-th token in the sequence, and \(\mathcal{V}\) is the vocabulary. \(M_{\text{orig}}\) denotes the original model without a watermark, \(M_{\text{wm}}\) is the watermarked model, and \(sk \in \mathcal{S}\) is the watermark secret key sampled from \(\mathcal{S}\).
Researchers have developed a groundbreaking system that uses bacteria to mimic the problem-solving capabilities of artificial neural networks.
Cell-based biocomputing is a novel technique that uses cellular processes to perform computations. Such micron-scale biocomputers could overcome many of the energy, cost and technological limitations of conventional microprocessor-based computers, but the technology is still very much in its infancy. One of the key challenges is the creation of cell-based systems that can solve complex computational problems.
Now a research team from the Saha Institute of Nuclear Physics in India has used genetically modified bacteria to create a cell-based biocomputer with problem-solving capabilities. The researchers created 14 engineered bacterial cells, each of which functioned as a modular and configurable system. They demonstrated that by mixing and matching appropriate modules, the resulting multicellular system could solve nine yes/no computational decision problems and one optimization problem.
The cellular system, described in Nature Chemical Biology, can identify prime numbers, check whether a given letter is a vowel, and even determine the maximum number of pizza or pie slices obtained from a specific number of straight cuts. Here, senior author Sangram Bagh explains the study’s aims and findings.
Mental health issues are one of the most common causes of disability, affecting more than a billion people worldwide. Addressing mental health difficulties can present extraordinarily tough problems: what can providers do to help people in the most precarious situations? How do changes in the physical brain affect our thoughts and experiences? And at the end of the day, how can everyone get the care they need?
Answering those questions was the shared goal of the researchers who attended the Mental Health, Brain, and Behavioral Science Research Day in September. While the problems they faced were serious, the new solutions they started to build could ultimately help improve mental health care at individual and societal levels.
“We’re building something that there’s no blueprint for,” said Mark Rapaport, MD, CEO of Huntsman Mental Health Institute at the University of Utah. “We’re developing new and durable ways of addressing some of the most difficult issues we face in society.”
Deep-learning models are being used in many fields, from health care diagnostics to financial forecasting. However, these models are so computationally intensive that they require the use of powerful cloud-based servers.
Google discontinues Artificial Intelligence and establishes DeepMind: It is capable of making microscopic predictions about the future.