News/Research
LLM runs on Commodore 64 in impressive display of 80s tech staying power
The 42-year-old Commodore 64 just became the oldest computer capable of running a large language model, showcasing the remarkable versatility of early computing hardware in the age of AI. While modern AI companies race to optimize their models for efficiency on contemporary devices, developer Maciej Witkowiak has taken a dramatically different approach by successfully porting a simplified LLM to run on 1982 technology, demonstrating how even the most basic computing platforms can participate in today's AI revolution. The big picture: Developer Maciej Witkowiak has successfully ported a simplified version of Llama 2 to run on a Commodore 64 computer from...
read May 21, 2025New NH prediction model transforms agricultural market forecasting in Korea
The NH Prediction project represents a significant advance in agricultural price forecasting for the Korean market, addressing the critical challenge of price volatility that affects all participants in the food supply chain. By leveraging 40 years of historical wholesale price data and applying 14 advanced prediction models, this research aims to bring unprecedented accuracy to agricultural market forecasting through customized algorithms that account for the unique characteristics of Korean agriculture, including seasonal variations and structural market changes. The big picture: Researchers have developed NH Prediction, an AI-powered forecasting system specifically designed to address price volatility in Korean agricultural and fishery...
read May 20, 2025The science behind diffusion models and AI image creation
Diffusion models represent a revolutionary approach to AI image generation, operating on fundamentally different principles than the transformer-based large language models dominating text generation. Unlike the relatively straightforward token prediction process of language models, diffusion models work by learning to reverse noise addition—essentially reconstructing images by progressively removing layers of random static. This conceptual framework has become nearly as influential in the AI revolution as transformers, enabling the remarkable image generation capabilities seen in modern AI systems. The core mechanism: Diffusion models work by understanding the gradient between pure noise and coherent images. The fundamental principle involves taking a clear...
read May 20, 2025AI benchmarks are losing credibility as companies game the system
As AI benchmarks gain prominence in Silicon Valley, they face increasing scrutiny over their accuracy and validity. The popular SWE-Bench coding benchmark, which evaluates AI models using real-world programming problems, has become a key metric for major companies like OpenAI, Anthropic, and Google. However, this competitive atmosphere has led to benchmark gaming and raised fundamental questions about how we measure AI capabilities. The industry now faces a critical challenge: developing more meaningful evaluation methods that accurately reflect real-world AI performance rather than just optimizing for test scores. The big picture: AI benchmarks like SWE-Bench have become crucial competitive metrics in...
read May 20, 2025AI impersonation scandal prompts Reddit to rethink anonymity
Reddit's plan to combat AI fraud on its platform marks a significant shift in how the social media site will verify user identity, potentially challenging its longstanding commitment to anonymity. The company's response follows an unethical AI experiment conducted without user consent, highlighting the growing tension between preserving authentic human interaction and maintaining user privacy in online communities. The unauthorized experiment: University of Zurich researchers conducted an extensive AI fraud operation in the popular Change My View subreddit, violating ethical standards and Reddit's policies. The researchers deployed AI bots that created over 1,700 comments while impersonating humans, including sensitive personas...
read May 20, 2025How Cosmos Reason helps AI understand and act in the real world
Cosmos Reason represents a significant advancement in physical AI, offering multimodal reasoning capabilities that bridge the gap between perception and decision-making in real-world contexts. This new world foundation model (WFM) combines video understanding with chain-of-thought reasoning, enabling it to understand physical common sense and make embodied decisions—capabilities that could transform how robots and autonomous vehicles learn to navigate and interact with their environments. The big picture: Cosmos Reason is designed not just to see but to reason about physical reality, processing both video and text inputs to generate thoughtful responses about real-world situations. The model demonstrates strong physical common-sense reasoning,...
read May 20, 2025LLMs like GPT-4 can outperform humans in online debates when given personal data, study finds
New research reveals that large language models (LLMs) outperform humans in persuasive online debates, especially when they can leverage personal data to tailor arguments. This breakthrough, detailed in Nature Human Behavior, demonstrates AI's growing capacity to change minds at scale—creating both promising applications for countering misinformation and concerning implications for digital influence campaigns that could reshape public opinion through personalized persuasion. The big picture: Researchers found OpenAI's GPT-4 was significantly more persuasive than humans in debates, particularly when given access to personal information about debate opponents. The multi-university study involved 900 US participants who provided personal details like age, gender,...
read May 20, 2025AI model offers breakthrough in early detection of postpartum depression
A new machine learning model shows promise in identifying parents at high risk for postpartum depression, potentially enabling targeted preventive care for those most vulnerable to this widespread condition. This predictive technology represents a significant step forward in maternal mental health care, addressing a condition that affects 17% of birthing parents globally and often goes undetected until symptoms become severe. The big picture: Researchers have developed an AI tool that can identify individuals with a high likelihood of developing postpartum depression, allowing healthcare providers to focus limited mental health resources where they're most needed. The model identified a high-risk group...
read May 20, 2025OpenAI offers $250K prize for AI-driven archaeological breakthroughs
OpenAI's new competition aims to uncover hidden archaeological sites in the Amazon rainforest through artificial intelligence, challenging conventional archaeological paradigms that have historically neglected this vast ecosystem. This initiative not only democratizes archaeological research by making it accessible to anyone with internet access but also validates indigenous oral traditions that have long claimed the Amazon was home to advanced civilizations. By harnessing AI to analyze diverse data sources from satellite imagery to colonial diaries, OpenAI is creating a digital treasure hunt with significant implications for our understanding of human history. The big picture: OpenAI has launched the "OpenAI to Z...
read May 20, 2025AI expands creative horizons with 4 new directions including audio and XR
Artificial intelligence continues to evolve beyond text-based applications into new creative domains, opening fresh possibilities for businesses and creators. A recent panel at an Imagination in Action event revealed how AI researchers and companies are expanding neural network technologies into audio processing, gaming analysis, extended reality, and biomimicry. These emerging directions highlight how AI is becoming increasingly multisensory and adaptive, requiring enterprises to understand both technological capabilities and the changing expectations of AI-native generations. 1. Mining the audio world Audio represents a largely untapped frontier for AI development compared to text-based systems. While text has been the primary focus of...
read May 20, 2025AI multipolarity gains importance in global tech landscape
The multipolar approach to AI development offers a compelling alternative to centralized control models, potentially creating more resilient, adaptable, and inclusive technological growth pathways. While current AI safety discussions often default to unipolar frameworks, exploring decentralized governance structures could address key risks like value lock-in and institutional stagnation while opening doors to more cooperative and human-empowering technological progress. The big picture: Multipolar AI scenarios envision a diverse ecosystem of AI agents, human actors, and hybrid entities cooperating through decentralized frameworks, in contrast to unipolar models that concentrate AI control under a single global authority. Key challenges: Multipolar AI development faces...
read May 20, 2025AI cyber defense hub emerges in Wolverhampton
Wolverhampton University's new research center marks a significant step in the UK's approach to cybersecurity and artificial intelligence by combining these two critical fields under one roof. As cyber attacks continue to increase across Europe and AI adoption accelerates at unprecedented rates, this initiative represents a strategic response to evolving digital threats and opportunities, focusing on resilience rather than just prevention. The big picture: The Centre for Cyber Resilience and Artificial Intelligence (CYBRAI) launched at Wolverhampton University combines two rapidly converging technological domains that are transforming business operations worldwide. The research hub brings together academia, industry, and government to tackle...
read May 20, 2025These AI tools from Meta could fast-track drug discovery and climate solutions
Meta AI is accelerating scientific progress through powerful new models and datasets for molecular discovery, potentially revolutionizing fields from drug development to climate solutions. Their breakthroughs include the largest quantum chemistry dataset ever created, a universal model for simulating atomic behavior, and a generative modeling technique that creates diverse molecules without requiring existing data. These innovations, combined with their neuroscience research revealing parallels between AI and human language development, showcase how advanced machine intelligence can transform scientific discovery by compressing decades-long experimental processes into rapid computational simulations. The big picture: Meta's FAIR research lab has released a suite of breakthrough...
read May 19, 2025Toronto AI safety startup Trajectory Labs launches
Trajectory Labs launches in Toronto to accelerate local AI safety initiatives through workspace, events, and community building. This nonprofit, funded by Open Philanthropy and operational since January 2025, aims to tap into Toronto's potential to contribute meaningfully to AI safety. By creating a physical hub for collaboration and knowledge sharing, Trajectory Labs is addressing the practical needs of the growing AI safety community in a tech-forward Canadian city, offering an important model for how local ecosystems can organize to tackle global AI challenges. The big picture: Trajectory Labs has established a nonprofit AI safety hub in downtown Toronto with funding...
read May 19, 2025LegoGPT model creates custom Lego sets for free in novel form of AI “buildout”
Carnegie Mellon researchers have developed LegoGPT, an innovative AI tool that transforms simple text descriptions into physics-tested, buildable Lego designs. This free, open-source system represents a significant advancement in AI-generated physical objects, offering step-by-step brick-by-brick instructions that bridge the gap between creative imagination and real-world construction. By combining generative AI with physics simulations, LegoGPT demonstrates how artificial intelligence can create designs that aren't just visually appealing but structurally sound and physically buildable. How it works: LegoGPT converts natural language descriptions into complete Lego building instructions that can be physically constructed using real bricks. The system employs physics simulations to test...
read May 19, 2025AI-designed irritable bowel disease drug enters human trials for the first time
Absci's generative AI-created antibody for irritable bowel disease enters clinical trials, marking a significant milestone for AI drug development. This Phase I trial represents one of the first attempts to secure regulatory approval for a therapy designed entirely using artificial intelligence, highlighting the pharmaceutical industry's ongoing efforts to transform the traditionally lengthy and expensive drug development process through AI-powered innovation. The big picture: Absci has begun Phase I clinical trials for ABS-101, an antibody therapy for irritable bowel disease created from scratch using generative AI. The company has officially dosed the first patients in the trial, administering the drug to...
read May 19, 2025Maryland university launches AI-powered education tech hub with cross-country partners
The University of Maryland has launched the Center for Educational Data Science and Innovation (EDSI), taking a research-first approach to developing AI-powered educational technology. Unlike many edtech startups rushing to market with AI tools, EDSI prioritizes building ethical infrastructure and conducting rigorous research before deploying solutions. This deliberate strategy represents a significant counterbalance in a field increasingly dominated by rapid AI tool development, emphasizing evidence-based innovation over speed to market. The big picture: EDSI aims to address persistent educational challenges through AI while prioritizing responsible development and research validation before releasing new tools. The center focuses on building infrastructure that...
read May 19, 2025AI medical advice improves, but adoption remains a challenge
OpenAI's latest research shows chatbots are improving at answering medical questions, but a critical gap remains between the artificial testing environment and real-world medical emergencies. The company's new HealthBench evaluation framework tests how well AI models can provide medical advice through text-based interactions, yet it doesn't address how humans might actually interpret or act on AI-generated medical guidance during emergencies. This distinction highlights a fundamental challenge in medical AI: technical performance in controlled settings doesn't necessarily translate to beneficial real-world outcomes. The big picture: OpenAI's HealthBench tests AI models on their ability to respond appropriately to medical questions, including emergency...
read May 19, 2025Whisper AI transcribes 10x faster with new Inference Endpoints
Hugging Face has launched a dramatically improved Whisper model deployment option on Inference Endpoints, delivering up to 8x faster performance for audio transcription services. This advancement makes powerful transcription capabilities more accessible and cost-effective, bringing enterprise-grade speech recognition within reach of more organizations through optimized open-source technology. The big picture: Hugging Face's new Whisper deployment leverages the open-source vLLM project to achieve substantial performance gains without sacrificing transcription quality. The solution specifically targets audio transcription efficiency using Whisper Large V3, which demonstrates nearly 8x improvement in real-time factor (RTFx) compared to previous versions. Word Error Rate (WER) evaluations across eight...
read May 19, 2025AI productivity study cited by Nature, Wall Street Journal retracted amid fraud allegations
A major AI productivity research paper claiming significant benefits for material scientists has been retracted amid serious fraud concerns. The study, which reported a 44% increase in materials discovery and 81% productivity boost for top scientists after implementing an AI tool, was widely covered in prestigious outlets and endorsed by Nobel laureate Darren Acemoglu. This case highlights the critical importance of scrutinizing AI research claims, especially as organizations make strategic decisions based on purported productivity improvements. The big picture: An influential paper claiming dramatic AI-driven productivity gains in scientific discovery has been withdrawn amid allegations of data fabrication and research...
read May 19, 2025AI unravels 3D genome structure mysteries
Scientists are making remarkable progress in understanding how 2 meters of DNA fits inside a microscopic cell nucleus and creates different cell types through varying gene expression patterns. MIT Professor Bin Zhang is pioneering computational approaches to tackle this complex biological challenge, using both computer simulations and generative AI to model 3D genome structures. By decoding these intricate chromatin structures—the combination of DNA and proteins that determines which genes are accessible for transcription—researchers hope to unlock fundamental insights into cellular diversity and potentially develop new therapeutic approaches. The big picture: MIT researchers are using artificial intelligence to solve a fundamental...
read May 19, 2025AI evaluation research methods detect AI “safetywashing” and other fails
The AI safety research community is making significant progress in developing measurement frameworks to evaluate the safety aspects of advanced systems. A new systematic literature review attempts to organize the growing field of AI safety evaluation methods, providing a comprehensive taxonomy and highlighting both progress and limitations. Understanding these measurement approaches is crucial as AI systems become more capable and potentially dangerous, offering a roadmap for researchers and organizations committed to responsible AI development. The big picture: Researchers have created a systematic literature review of AI safety evaluation methods, organizing the field into three key dimensions: what properties to measure,...
read May 19, 2025AI chips evolve as imec lab pursues programmable designs
The rapid evolution of artificial intelligence algorithms is creating an unexpected challenge for chipmakers, with traditional design approaches potentially becoming a bottleneck for future AI development. According to Imec CEO Luc Van den hove, the semiconductor industry must pivot toward more flexible, reconfigurable architectures to keep pace with AI's swift advancement, or risk creating costly hardware that becomes obsolete before it's even deployed. The big picture: One of the world's leading semiconductor R&D organizations warns that current chip design strategies focusing on raw computing power are creating unsustainable energy consumption, rising costs, and slow hardware development cycles. "There is a...
read May 19, 2025CausVid enables interactive AI video generation on the fly
MIT researchers have developed a new AI video generation approach that combines the quality of full-sequence diffusion models with the speed of frame-by-frame generation. Called "CausVid," this hybrid system creates videos in seconds rather than through the slow, all-at-once processing used by models like OpenAI's SORA and Google's VEO 2. This breakthrough enables interactive, on-the-fly video creation that could transform various applications from video editing to gaming and robotics training. The big picture: MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) and Adobe Research have created a video generation system that works like a student learning from a teacher, where...
read