AI News 3/22/2024
Scientists create AI models that can talk to each other and pass on skills with limited human input
Researchers have developed AI models capable of human-like communication and knowledge transfer with minimal human intervention. These models can teach each other tasks using natural language processing (NLP), based on written instructions alone. The first AI, equipped with an NLP model known as "S-Bert," learns a task and then instructs a second, inexperienced AI to perform the same task successfully, marking a significant advancement in AI communication. This breakthrough, published in the journal Nature, not only enhances the understanding of human cognitive functions but also holds potential for applications in robotics and automated systems where AIs can efficiently transfer knowledge among themselves. The development is a step toward more complex networks and humanoid robots that can understand and interact with humans and each other, underscoring the evolving capabilities of AI in simulating human learning and communication processes.
GTC March 2024 Keynote with NVIDIA CEO Jensen Huang - Project GROOT
OpenAI'S Q-STAR Has More SHOCKING LEAKED Details! (Open AI Q*)
Anon Leaks NEW Details About Q* | "This is AGI"
7 Incredible Al Tools You’ve DEFINITELY Never Seen Before! (Underground AI #2)
Vernor Vinge Passes Away
Vernor Vinge, a visionary science fiction author and academic, has passed away at the age of 79 due to complications from Parkinson's Disease. Vinge was renowned for his contributions to the science fiction genre, notably for popularizing the concept of the "Technological Singularity"—a point at which technological growth becomes uncontrollable and irreversible, resulting in unforeseeable changes to human civilization. He explored this theme and others through his work, earning acclaim and influencing both the literary world and technological leaders. As a professor of mathematics and computer science at San Diego State University until his retirement in 2000, Vinge not only captivated audiences with his narratives but also shaped future discussions on artificial intelligence and the potential trajectories of human progress. His predictions and literary achievements, including Hugo Award-winning novels, underscore his profound impact on how we envision the future of humanity intertwined with technology.
Source: https://www.popsci.com/science/vernor-vinge-obit/
Method identified to double computer processing speeds
Researchers at the University of California - Riverside have developed a new computer processing method called "Simultaneous and Heterogeneous Multithreading" (SHMT), which can double computer processing speeds using existing hardware. This approach utilizes graphics processing units (GPUs), AI and machine learning hardware accelerators, and digital signal processing units in tandem to process information, overcoming the bottleneck of moving information between separate units. The proposed SHMT framework, tested on an embedded system, achieved a 1.96 times speedup and reduced energy consumption by 51%. This breakthrough could lower computer hardware costs, reduce carbon emissions from data centers, and decrease the freshwater needed for cooling servers. Further research is needed to explore system implementation, hardware support, and application benefits.
Source: https://www.sciencedaily.com/releases/2024/02/240221213907.htm
Two artificial intelligences talk to each other
Researchers at the Université de Genève have created an artificial neural network capable of learning tasks from verbal or written instructions and then describing these tasks linguistically to another AI, which can then perform them. This development, significant especially for robotics, demonstrates an advanced level of cognitive capability in AI, resembling human-like communication and learning abilities. The study, published in Nature Neuroscience, marks a pioneering step towards enabling AI systems to understand and execute tasks based purely on language instructions, a domain previously challenging for AI technology. For more details, you can read the full article on ScienceDaily.
Source: https://www.sciencedaily.com/releases/2024/03/240318142438.htm
Researchers Give AI an Inner Monologue
Researchers developed an AI, Quiet-STaR, that "thinks" before answering, simulating a human's inner monologue. Built on the Mistral 7B model, it was designed to reason and improve its responses by showing its "thought process." Users then selected the most accurate responses, helping the AI refine its reasoning skills. Though its accuracy in tasks like math remains low, Quiet-STaR's ability to double its math performance and demonstrate improved reasoning capabilities offers promising steps toward bridging the gap between current AI reasoning and human-like reasoning. For more details, you can read the full article on Futurism.
Source: https://futurism.com/the-byte/ai-inner-monologue
10% of US Workers are in Roles At Risk From AI Developments
A White House analysis indicates that 10% of US workers are in roles most at risk from AI developments, particularly affecting those with lower education and income, potentially exacerbating inequality. The analysis, part of the Economic Report of the President, underscores the importance of preparing for AI's impact on the workforce. It suggests that while AI might alter jobs by automating certain tasks, it's not necessarily expected to eliminate them entirely, highlighting the nuanced potential effects on employment.
Source: https://www.cnn.com/2024/03/21/business/ai-jobs-white-house-work-force/index.html
Tennessee Passes ELVIS Law To Protect Musicians
Tennessee has become the first U.S. state to legislate protections for musicians and artists against unauthorized AI impersonation with the ELVIS Act. This groundbreaking law, effective from July 1, extends the state's right of publicity law to include AI-specific protections, prohibiting the use of AI to mimic an artist's voice without permission. This move, endorsed by music stars and the governor, aims to safeguard the creative integrity and copyrights of artists amid the rapid advancement of AI technology in the music industry. For more details, visit NPR.
Source: https://www.npr.org/2024/03/22/1240114159/tennessee-protect-musicians-artists-ai
AI Tool Finds Cancer Signs Missed By Doctors
An AI tool named Mia, developed for use in the NHS, has demonstrated the ability to detect signs of breast cancer in mammograms that were missed by human radiologists. During a trial involving over 10,000 women, Mia not only identified all participants with symptoms of breast cancer but also found an additional 11 cases that doctors had overlooked. With an accuracy rate of 81.6% in predicting cancer presence, Mia represents a significant advancement in cancer detection, potentially improving early diagnosis and treatment outcomes. Researchers are now aiming to enhance the system further for predicting patient risk of post-treatment side effects. For more details, visit Artificial Intelligence News.
Source: https://www.artificialintelligence-news.com/2024/03/21/ai-tool-finds-cancer-signs-missed-by-doctors/
AI generates high-quality images 30 times faster in a single step
MIT researchers have developed a new AI framework that generates high-quality images 30 times faster than current methods, reducing the image-generating process to a single step. This method, which simplifies traditional multi-step diffusion models, could significantly enhance design tools and potentially aid in drug discovery and 3D modeling. The breakthrough suggests a promising future for rapid content creation and high-quality real-time visual editing.
Source: https://news.mit.edu/2024/ai-generates-high-quality-images-30-times-faster-single-step-0321
AI Term of the Day: Convolutional neural network
A convolutional neural network (CNN) is a regularized type of feed-forward neural network that learns feature engineering by itself via filters (or kernel) optimization. Vanishing gradients and exploding gradients, seen during backpropagation in earlier neural networks, are prevented by using regularized weights over fewer connections. For example, for each neuron in the fully-connected layer 10,000 weights would be required for processing an image sized 100 × 100 pixels. However, applying cascaded convolution (or cross-correlation) kernels, only 25 neurons are required to process 5x5-sized tiles. Higher-layer features are extracted from wider context windows, compared to lower-layer features.
Source: https://en.wikipedia.org/wiki/Convolutional_neural_network
(image generated by DALL-E on ChatGPT-4)