- NATURAL 20
- Posts
- DARPA's AI Dogfight: F-16 vs. Human
DARPA's AI Dogfight: F-16 vs. Human
PLUS: Adobe Enhances AI with Firefly Image 3, Perplexity AI Hits $1 Billion Valuation and more.
Today:
DARPA's AI Dogfight: F-16 vs. Human
OpenAI's New Enterprise Grade for API customers
OpenCRISPR-1 Unveils, AI Enhances Genetics
Microsoft Launches Phi-3 Model
Adobe Enhances AI with Firefly Image 3
Perplexity AI Hits $1 Billion Valuation
Multimodal AI Powers Ray-Ban Meta Glasses
DARPA BOMBSHELL "AI-piloted F-16 engaging in dogfights against humans" | ANDURIL, Midjourney Random
The U.S. Air Force has successfully conducted tests with AI-controlled aircraft at Edwards Air Force Base, California. An AI agent piloted an X-62 VISTA in a simulated dogfight against a human-piloted F-16, demonstrating the capability for safe, autonomous air combat. This milestone was part of DARPA’s Air Combat Evolution program, aimed at developing AI-driven autonomy for air combat.
The tests showed rapid progress, moving from AI installation to live demonstration within a year, and involved extensive software updates across multiple test flights. Safety pilots were present but intervention was unnecessary, marking significant advancements in autonomous military technologies.
OpenAI shrugs off Meta’s Llama 3 ascent with new enterprise AI features
OpenAI isn't sweating Meta's Llama 3's rise in AI popularity. Instead, it's upping its game with new features for its GPT-4 Turbo model. OpenAI's latest move includes beefing up security with Private Link and Multi-Factor Authentication. Their Assistants API now handles 500 times more files, making it a powerhouse for custom models. Projects allow organizations to partition tasks securely.
Plus, new cost management features offer discounts and faster results for batch processing. OpenAI aims for a more user-friendly enterprise AI experience, countering competitors like Llama 3 and Mistral.
Generative A.I. Arrives in the Gene Editing World of CRISPR
A fresh A.I. breakthrough shakes up gene editing, akin to ChatGPT's poetry prowess. Profluent's tech, inspired by ChatGPT's methods, crafts DNA editing blueprints, promising a swift, precise healthcare future. Leveraging CRISPR, it mimics nature's mechanisms but births new ones. OpenCRISPR-1, a product of this innovation, opens a new era, free for all to tinker with.
Can AI rewrite our human genome? ⌨️🧬
Today, we announce the successful editing of DNA in human cells with gene editors fully designed with AI. Not only that, we've decided to freely release the molecules under the @ProfluentBio OpenCRISPR initiative.
Lots to unpack👇
— Ali Madani (@thisismadani)
8:43 PM • Apr 22, 2024
Though it's early days, the potential is vast, hinting at personalized treatments and quicker medical breakthroughs. Yet, ethical concerns linger, cautioning against misuse. While the road ahead is uncertain, the possibilities are vast, ushering in a new frontier in healthcare.
Microsoft Phi-3: Tiny language model, huge implications for enterprise AI adoption
Microsoft unveils Phi-3, a compact 3 billion parameter language model rivaling larger counterparts like GPT-3.5 in capabilities but at a fraction of the cost. Following Phi-1's debut, the Phi series progresses with notable achievements in coding and reasoning tasks. Azure AI now hosts Phi-3, empowering businesses with cost-effective natural language processing. Eric Boyd emphasizes Azure's role in enabling generative AI applications while maintaining affordability.
Phi-3's development prioritizes responsible AI principles, ensuring data integrity and safety. Optimized for NVIDIA GPUs, Phi-3 boasts efficient attention mechanisms, promising high performance on commodity hardware. Its launch signifies a leap in democratizing large language models, essential for AI-driven business innovation.
Adobe Photoshop gets new AI image generation features with Firefly Image 3 upgrade
Adobe Photoshop is upgrading with Firefly Image 3, introducing a new Generate Image feature. This AI-driven enhancement promises higher-quality image generation, better detail, and variety. Notably, it's trained on licensed and public domain content, ensuring copyright compliance. The new Generate Image feature creates images from text prompts, setting it apart from existing features like Generative Fill.
Generative Fill also receives upgrades, including control over outputs, background editing, creating variations, and enhancing detail. These advancements aim to streamline creative processes in Photoshop, garnering significant user interest. Are you excited about these AI additions to Photoshop? Share your thoughts!
AI Search Startup Perplexity Valued at $1 Billion in Funding Round
Perplexity AI, a startup rivaling Google with its AI-powered search engine, secures $63 million in funding, valuing the company at over $1 billion. Led by investor Daniel Gross, backers include luminaries like Stanley Druckenmiller and Jeff Bezos. Founded less than two years ago, Perplexity offers an AI chatbot that delivers accurate search results and has gained traction, processing 75 million queries in the US this year.
With a focus on precision, Perplexity targets businesses with a new enterprise version priced at $40/month. Despite competition from tech giants, Perplexity's agile approach and accuracy stand out in the evolving AI landscape.
The Ray-Ban Meta Smart Glasses have multimodal AI now
The Ray-Ban Meta Smart Glasses, now equipped with multimodal AI, offer a blend of convenience and occasional quirks. Initially lacking this feature, they've since caught up, allowing users to process various types of information like photos, audio, and text. Despite the recent flop of the Humane AI Pin, these glasses show promise. While not flawless, they excel in tasks like identifying objects and providing information.
However, they have limitations, like difficulty in zooming. Yet, their integration with familiar tech like smartphones and their comfortable design make them practical for everyday use, especially outdoors. Overall, they represent a step towards mainstream acceptance of wearable AI.
🧠RESEARCH
Phi-3 Technical Report introduces phi-3-mini, a 3.8 billion parameter language model trained on 3.3 trillion tokens, performing comparably to larger models like Mixtral 8x7B and GPT-3.5. It's designed for phone deployment, trained on a scaled-up dataset from phi-2, emphasizing robustness, safety, and chat format. Larger models, phi-3-small and phi-3-medium, show even greater capabilities.
The paper discusses vulnerabilities in Language Models (LLMs) like GPT-3.5 due to prompt injections and jailbreaks, allowing malicious prompts. They propose an instruction hierarchy to prioritize instructions, aiming to ignore lower-priority ones. Testing on GPT-3.5 shows increased robustness against attacks, with minimal impact on standard abilities.
The paper assesses the performance of Meta's LLaMA3 models when quantized to low bit-widths, crucial for resource-limited settings. Despite impressive performance of LLaMA3, quantization to 1-8 bits reveals significant degradation, especially at ultra-low bit-widths. The study underscores the need for bridging performance gaps in low-bit quantization for future LLM developments, aiding practical applications.
The paper introduces Hyper-SD, a novel framework for efficient image synthesis, addressing limitations in current diffusion model distillation methods. Hyper-SD synergistically combines trajectory preservation and reformulation, maintaining near-lossless performance during step compression. It incorporates human feedback learning and score distillation, achieving state-of-the-art performance across inference steps.
The paper presents FlowMind, an automatic workflow generation system utilizing Large Language Models (LLMs) like GPT. FlowMind addresses limitations in Robotic Process Automation (RPA) by enabling spontaneous task handling. It employs a generic prompt recipe grounded in reliable APIs, mitigating hallucinations and ensuring data integrity. FlowMind simplifies user interaction and feedback, demonstrating success in finance tasks with NCEN-QA dataset.
🛠️TOP TOOLS
Melodisco - Play AI music.
Guidde - generative AI platform for business that helps your team create video documentation 11x faster
Noctie AI - Improve at chess – learn from a humanlike chess AI
Flash Notes - Transform your learning with AI-enhanced, personalised flashcards and seamless device synchronisation
Autolab - Automate tedious tasks using AI
📲SOCIAL MEDIA
Exciting update -- Llama-3 full result is out, now reaching top-5 on the Arena leaderboard🔥
We've got stable enough CIs with over 12K votes. No question now Llama-3 70B is the new king of open model. Its powerful 8B variant has also surpassed many larger-size models. What an… twitter.com/i/web/status/1…
— lmsys.org (@lmsysorg)
6:56 PM • Apr 22, 2024
🗞️MORE NEWS
Mabl launches an AI-powered low-code mobile app testing service
Mabl introduces AI-driven mobile app testing, simplifying development with low-code solutions. Their unified platform promises to streamline testing across various interfaces, including APIs, browsers, and devices. By leveraging machine learning and computer vision, Mabl aims to tackle challenges like manual testing and deployment complexities, offering transparent pricing tailored to organizational needs. VENTUREBEAT
Meta AI adds Google Search results
Meta AI now incorporates Google Search results alongside Bing's, a first in AI assistance. This integration aims to provide users with the most current answers. As AI-powered assistants reshape search habits, Meta's move acknowledges the importance of real-time data access, enhancing user experience across Facebook's platforms. SEARCH ENGINE LAND
Vietnam's FPT to invest $200 million in AI factory using Nvidia chips
Vietnam's leading tech company, FPT, and Nvidia, a U.S. chip giant, announced a $200 million investment in an AI factory. The partnership aims to enhance AI research in Vietnam, focusing on generative AI and autonomous driving. FPT plans to expand operations to Japan and South Korea, aiming to establish Vietnam as an AI hub. YAHOO!
SimpliSafe’s new outdoor monitoring service combines AI with live agents
SimpliSafe introduces outdoor live monitoring with facial recognition, combining AI with human agents. The system identifies potential intruders and allows agents to take immediate action, such as warning or dispatching help. The service, part of an Early Access Program, expands proactive security measures beyond traditional alarm systems. THE VERGE
Brev.dev joins forces with Akash Network to offer on-demand GPUs for AI developers
Brev.dev partners with Akash Network, enabling AI developers to access GPUs via a decentralized cloud marketplace. This collaboration addresses the challenges of limited computing resources and high costs, empowering developers to scale AI projects affordably. With on-demand GPU access, the partnership aims to democratize AI development and drive innovation in the industry. VENTUREBEAT
AI tool creates 'synthetic' images of cells for enhanced microscopy analysis
UC Santa Cruz researchers have developed an AI tool called cGAN-Seg, which generates synthetic images of single cells to aid in microscopy analysis. These images, resembling real cells, help train AI models for accurate cell segmentation, improving research efficiency in studying cell behavior and diseases. PHYS.ORG
AI Developers Agree to New Safety Measures to Fight Child Exploitation
Major AI companies, including OpenAI, Meta, and Google, commit to new safety measures to combat child exploitation online. The initiative aims to curb the creation of exploitative content, leveraging generative AI tools, in collaboration with child safety organizations. This move underscores efforts to protect vulnerable individuals in the digital space. THE WALL STREET JOURNAL
What'd you think of today's edition? |
What are MOST interested in learning about AI?What stories or resources will be most interesting for you to hear about? |
Reply