News/Science
AI Discovers Rare-Earth-Free Magnet, Promising Cheaper, Greener EVs and Tech
UK company Materials Nexus has used its AI platform to discover a new rare-earth-free permanent magnet called MagNex, 200 times faster than traditional methods and with the potential to significantly reduce material costs and carbon emissions in magnet production. Growing demand for rare earth magnets in electric motors: The shift towards electric vehicles is driving a rapid rise in demand for compact, high-power motors that rely on rare earth magnets, with Materials Nexus estimating a tenfold increase in permanent magnet demand by 2030 in the EV industry alone. Over 80% of modern electric vehicles use permanent magnet motors, which are...
read Jun 25, 2024AI Breakthrough: Simulating Millions of Years of Protein Evolution in Days
A former Meta engineer's AI research lab, EvolutionaryScale, has launched ESM3, a multimodal generative language model that can design novel proteins, simulating millions of years of evolution in a matter of days. Groundbreaking capabilities: ESM3 is a frontier generative model for biology that can reason across the fundamental properties of proteins - sequence, structure, and function: The model was trained on 2.78 billion natural proteins and 771 billion unique tokens, enabling it to generate novel proteins with an unprecedented level of control. In tests, ESM3 generated a novel green fluorescent protein (esmGFP) that would have taken over 500 million years...
read Jun 21, 2024AI Experts and Hackers Unite to Expose Dangers of Advanced AI Models
A coordinated effort by AI experts and "hackers" aims to expose potential dangers posed by powerful AI models, highlighting the need for robust safeguards as the technology rapidly advances. Key Takeaways: The "Jailbreak Challenge" seeks to reveal vulnerabilities in AI models and spur the development of stronger safety measures: Organized by Anthropic, the initiative involves over 3,000 AI experts and "hackers" attempting to manipulate AI assistants into violating their programming or engaging in harmful activities. Early results show that even advanced AI models like ChatGPT can be tricked into bypassing safety restrictions, generating content related to crime, violence, and explicit...
read Jun 21, 2024Ray Kurzweil: AI-Human Merger by 2045 Inevitable, Fears Delay Progress
Renowned futurist Ray Kurzweil doubles down on his prediction of AI-human merger by 2045 in his new book, arguing that focusing on AI's potential dangers could delay progress and prolong human suffering. The key points and implications are: Kurzweil's steadfast vision: Kurzweil reaffirms his bold prediction from 2005 that AI will reach human-level intelligence by 2029 and merge with humans by 2045, an event he calls "The Singularity": He believes exponential growth in computing power and falling costs make this merger inevitable, expanding human consciousness in unimaginable ways via brain-computer interfaces. Kurzweil expects AI to have the most profound near-term...
read Jun 21, 2024Breakthrough AI Identifies Tumors, Outperforming Top Models like ChatGPT-4
A new pathology-specific large language model, PathChat, demonstrates breakthrough capabilities in identifying and diagnosing tumors, outperforming leading AI models like ChatGPT-4 and LLaVA. Key Takeaways: PathChat represents a significant advancement in computational pathology, serving as an AI copilot for human pathologists: PathChat correctly identified the location and potential severity of a malignant eye tumor, while other state-of-the-art models failed to do so accurately. The model performed with 78% accuracy when presented with medical images alone and 89.5% accuracy when provided with additional clinical context, surpassing the performance of ChatGPT-4, LLaVA, and LLaVA-Med. PathChat's ability to adapt to downstream tasks like...
read