- NATURAL 20
- Posts
- DARPA talks about AGI, GPT-5 and Gemini
DARPA talks about AGI, GPT-5 and Gemini
PLUS: Grok-1.5 Challenge to GPT-4's Dominance, AI Robots Bringing Personal Touch to Letters and more.
Wes Roth
April 01, 2024
Today:
DARPA talks about AGI, GPT-5 and Gemini
DeepMind's Breakthrough in Superhuman AI Fact-Checking
OpenAI's Voice Engine Transforms Text with a 15-Second Clip
Grok-1.5 Challenge to GPT-4's Dominance
AI Robots Bringing Personal Touch to Letters
DARPA's STUNNING AGI BOMBSHELL | AGI Timeline, Gemini plus search, OpenAI's GPT-5 & AI Cyber Attacks
Jimmy Apples shared insights from a DARPA Q&A in November 2023, highlighting the uneven progress across AI research areas. The document from DARPA discusses the integration of the planning piece into the Gemini model within Large Language Models, but admits to a lack of transparency on developments. It addresses DARPA's interface between traditional hardware and AI, noting ongoing software-hardware interface projects.
Concerns about quantum computing's impact on cybersecurity are mentioned, with DARPA involved in the QUET program for quantum-enhanced networking security, but not in quantum safe security directly.
Google DeepMind unveils ‘superhuman’ AI system that excels in fact-checking, saving costs and improving accuracy
Google DeepMind's research team has developed an AI system named SAFE, which surpasses human fact-checkers in accuracy and cost-efficiency. SAFE breaks down texts into facts, using Google Search to verify each. In tests, SAFE matched human accuracy 72% of the time and was preferred in 76% of disputes over human judgment.
Despite its success, questions about what constitutes "superhuman" ability arose, with critics arguing that SAFE's comparison should be with expert fact-checkers, not average workers. The system's low cost makes it appealing for handling the growing volume of information online. Its effectiveness across several language models highlights its potential in reducing misinformation, though calls for more transparency and rigorous testing against expert human standards emphasize the need for careful evaluation of AI in critical tasks like fact-checking.
OpenAI’s voice cloning AI model only needs a 15-second sample to work
OpenAI has launched a new text-to-voice technology named Voice Engine, capable of creating a synthetic voice from just a 15-second voice clip. This innovation allows the AI to read out text in the original speaker's voice or translate and speak in other languages. Initially, access to Voice Engine is limited to specific companies in various sectors, including education and health, to explore beneficial applications and ensure responsible use.
OpenAI developed this model using diverse data sources and has implemented measures such as explicit consent requirements, watermarking, and usage monitoring to prevent misuse. The technology has already been integrated into ChatGPT's Read Aloud feature. OpenAI also suggests broader protective steps like phasing out voice authentication and increasing public awareness on AI deepfakes to mitigate potential risks associated with voice cloning technologies.
Elon Musk announces Grok-1.5, nearing GPT-4 level performance
Elon Musk's xAI has introduced Grok-1.5, an upgrade to their Grok-1 AI model, aiming to rival the performance of leading language models like OpenAI's GPT-4. Grok-1.5 boasts improved reasoning and problem-solving skills, with significant advancements in understanding and generating content based on complex prompts. It performed notably well in benchmarks, achieving high scores in math and coding tasks and showing a marked improvement over Grok-1 in language understanding.
Although it still trails behind some of its competitors in certain areas, Grok-1.5's enhanced capabilities demonstrate xAI's commitment to evolving AI technology. Musk hints at an even more advanced model, Grok-2, currently in development, promising to surpass existing AIs across all metrics. Grok-1.5 will soon be available for early testers and users of the X platform's Grok chatbot, indicating a phased rollout that seeks to expand access and introduce new features over time.
AI-operated Robots Write Letters for Humans
The resurgence of letter writing has taken a modern twist with the introduction of AI-operated robots capable of handwriting notes for humans. Businesses and non-profits are turning to companies like Handwrytten to use robots for personalized letters and thank-you notes, aiming to forge stronger connections with their recipients. This tech echoes a long history of automated writing, dating back to Thomas Jefferson's fascination with the polygraph, a device that allowed him to duplicate his letters.
Today's handwriting technology, leveraging advancements like 3D printing and laser cutting, can also assist in crafting messages with AI-generated content, demonstrating humanity's enduring desire to communicate through written words, regardless of the medium.
🧠RESEARCH
AniPortrait is a new method to create realistic animations of faces from audio and a single photo. It uses 3D models and 2D facial landmarks derived from sound to animate faces in a natural and diverse manner. This technique promises better visual quality in animations and could be useful for editing facial motions and reenacting faces.
"Octree-GS" proposes a solution for real-time rendering challenges faced by 3D Gaussian splatting (3D-GS) in complex scenes. Addressing bottlenecks and inconsistent speeds, it introduces level-of-detail (LOD) structures, dynamically selecting resolutions for rendering fidelity and performance. This approach ensures efficient rendering without sacrificing detail.
"VP3D" enhances text-to-3D generation by leveraging 2D visual prompts to refine the process. Unlike previous methods relying solely on text prompts, VP3D utilizes visual prompts to improve model fidelity, addressing issues like unrealistic textures and inconsistency. Through experiments, VP3D demonstrates superior visual quality and enables stylized text-to-3D generation.
"RakutenAI-7B" presents large language models tailored for Japanese, outperforming other open 7B models on Japanese LM Harness benchmarks. The suite includes foundation, instruction-tuned, and chat-tuned models, released under the Apache 2.0 license. This advancement facilitates improved natural language processing for Japanese, benefiting various applications.
"TRIP" introduces a novel approach for image-to-video generation, addressing challenges in maintaining coherence and fidelity. Leveraging image noise prior, it facilitates inter-frame relational reasoning and temporal coherence via temporal residual learning. Through experiments on various datasets, TRIP demonstrates effectiveness in generating high-quality videos from static images.
🛠️TOP TOOLS
CB Insights - Generate comprehensive reports instantly with AI.
DryMerge - Automate Work with Plain English
Vetted - AI shopping agent for people who want the best quality products for their money
Weavely - Build and design forms, with incredible speed
MagicForm - Supercharge your top of funnel with a 24/7 AI that qualifies, converts, and follows up with leads.
🗞️MORE NEWS
Google revamps Maps to feature top restaurant, travel tips from NYT, Lonely Planet, Infatuation … and AI
Google Maps is enhancing its service to include restaurant and travel suggestions from New York Times, Lonely Planet, and The Infatuation, alongside AI-generated advice. Initially available in 40 US and Canadian cities, this update aims to make Maps a comprehensive travel and dining guide. Google also plans to incorporate contributions from smaller bloggers and is exploring ways to make the platform beneficial for content creators, without direct payment for their contributions. FORTUNE
Openstream.ai Bridges Human-Machine Conversations With Next-Gen Voice Agents
Openstream.ai introduces a breakthrough in conversational AI with its Enterprise Virtual Assistant (EVA) platform, offering businesses advanced voice agents and AI avatars capable of natural, human-like conversations. Unlike traditional AI that relies on scripts and can produce unrealistic "hallucinations," Openstream employs a neuro-symbolic approach, blending large language models with symbolic AI for nuanced dialogue management. PYMNTS
U.S. updates export curbs on AI chips and tools to China
The U.S. government has updated its export control rules to further restrict China's access to American artificial intelligence (AI) chips and chipmaking tools. This move is part of a broader strategy to impede Beijing's semiconductor industry due to national security concerns. The revised regulations specifically target advanced AI chips, like those designed by Nvidia, to prevent their use in China's growing tech sector, which the U.S. fears could strengthen China's military capabilities. Additionally, the new rules clarify that laptops containing restricted chips are also banned from shipment to China. REUTERS
Seattle’s Soundry AI creates text-to-sample generator to help prompt enhanced music creation
Seattle's Soundry AI is pioneering the fusion of generative artificial intelligence with music creation. Their innovative text-to-sound generator, developed as part of their participation in the Y Combinator startup accelerator, enables musicians to craft custom samples from textual descriptions, enhancing creativity and expression in music production. Co-founded by Diandre Ruiz, Mark Buckler, and Justin Parus, Soundry AI stands out in the competitive field of music technology by providing a unique solution that bypasses the limitations of traditional sample libraries and copyright issues. GEEKWIRE
Sepsis blood test combined with AI could offer early detection tool
Researchers at Lund University have developed a method combining a blood test and AI to identify sepsis early, potentially saving lives. This approach, effective in predicting severe complications, will be presented at a congress in Barcelona. It signifies a major step forward in diagnosing and treating sepsis, aiming to prevent organ failure and death through quicker, more accurate detection. THE GUARDIAN
DrugGPT: new AI tool could help doctors prescribe medicine in England
Oxford University has developed DrugGPT, an AI tool designed to improve drug prescription accuracy in England. By providing a safety net and detailed medication information, DrugGPT aims to reduce the 237 million medication errors occurring annually, which cost lives and £98 million. It supports doctors with instant recommendations, potential side effects, and drug interactions, backed by current research and guidelines. THE GUARDIAN
US, Japan to call for deeper cooperation in AI, semiconductors, Asahi says
Japan and the United States are planning to enhance their cooperation in advanced technology sectors, including artificial intelligence (AI) and semiconductors. This decision is set to be officially announced in a joint statement during Prime Minister Fumio Kishida's visit to the U.S. on April 10, where he will meet President Joe Biden. The collaboration aims to establish a new framework for AI research and development with participation from leading tech companies like Nvidia, Arm, and Amazon. REUTERS
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? |
Join the conversation