[AI News] OpenAI’s New Structure, AI for Human Motion, Sonus-1 LLMs, and Meta’s AI Integration Push
Hello Friends,
Earlier this week, I shared a post about an LLM being used in CAD design—a clear reminder of how versatile these models truly are. For many, LLMs are synonymous with chatbots, but their potential reaches far beyond that. From sentiment analysis to workflow automation, these models are redefining what’s possible across industries.
Even more exciting is the flexibility of the transformer architecture behind LLMs. Researchers are adapting it to new modalities like interpreting human gestures and reshaping fields like robotics through learning from physical interactions (read more below). This adaptability solidifies AI’s role as a general-purpose technology, akin to electricity or the personal computer—transforming industries in profound ways.
Enjoy this week’s AI update!
- Manny
[OPENAI]
OpenAI Announces Plans to Restructure for Mission Advancement 🚀
The Recap: OpenAI's Board of Directors is evaluating changes to its corporate structure to better support its mission of ensuring artificial general intelligence (AGI) benefits humanity. The proposed changes aim to strengthen both the non-profit and for-profit arms while enabling greater capital raising capabilities.
Highlights:
OpenAI began in 2015 as a non-profit research lab focused on AGI development, initially funded by donations including from Elon Musk
In 2019, the organization created a hybrid structure with a for-profit arm controlled by the non-profit to raise necessary capital ($10B+ estimated needed)
ChatGPT launched in 2022, now serving over 300 million weekly users, mostly for free
In 2024, OpenAI discovered a new research paradigm with o-series models showing enhanced reasoning capabilities
The company plans to transform its for-profit arm into a Delaware Public Benefit Corporation (PBC) to raise capital with conventional terms
The restructuring would make the non-profit one of the best-resourced in history through shares in the PBC
Competition in AI has intensified with hundreds of billions being invested by major companies
Key Takeaways: OpenAI's proposed restructuring reflects the massive capital requirements of advanced AI development and the need to balance mission-driven goals with practical business considerations. The shift to a PBC structure, already used by other AI companies, aims to secure necessary funding while maintaining the organization's commitment to ensuring AGI benefits humanity. The non-profit's enhanced resources through PBC shares could enable expanded charitable initiatives in healthcare, education, and science. → Read more here.
[MULTI-MODAL LLM]
The Language of Motion: Teaching AI to Understand and Create Natural Human Movement 🤖✨
The Recap: Stanford researchers have created an AI system that can understand and generate natural human movements by learning from speech, text, and motion. Like a skilled performer who can both understand stage directions and act them out naturally, this AI can take verbal instructions or speech and translate them into realistic human movements, or watch movement and understand the emotions behind it.
Highlights:
First AI system that can seamlessly work with both spoken instructions and physical movements
Can generate natural gestures that match someone's speech, like a virtual actor or presenter
Understands complex movement requests like "walk in a circle while talking" or "sit down while gesturing"
Creates more natural and expressive movements than previous systems
Can identify emotions just by watching how someone moves
Needs less training data than earlier systems to learn new movements
Breaks down human movement into natural parts (face, hands, upper body, lower body) for better understanding
Can edit movements piece by piece - like changing just the walking pattern while keeping hand gestures
Learns general movement patterns that work across different people and situations
Maintains natural-looking movements even when combining different actions like walking and talking
Key Takeaways: This research brings us closer to AI that truly understands human movement and communication. The system's ability to work with both words and actions opens new possibilities for video games, virtual reality, and digital avatars. By needing less training data and producing more natural movements, it makes realistic motion generation more accessible for various applications. → Read more here.
[NEW LLM]
Sonus-1: A New Family of AI Language Models Claims Top Performance 🏆📚
The Recap: Sonus AI has announced their Sonus-1 family of language models, claiming breakthrough performance across multiple benchmarks including reasoning, math, and coding. The company positions their flagship Sonus-1 Pro with Reasoning as a top competitor in the AI model landscape, achieving scores that would place it among leading models.
Highlights:
New model family includes four versions: Mini, Air, Pro, and Pro with Reasoning
Pro model with Reasoning claims exceptional benchmark scores:
90.15% on MMLU (general knowledge)
91.8% on MATH-500 (mathematics)
90.0% on HumanEval (coding)
97.0% on GSM-8k (math problem solving)
Each model targets different use cases:
Mini focuses on speed and cost-effectiveness
Air balances performance with resource usage
Pro versions aim for maximum capability
Models are publicly accessible through chat.sonus.ai
Company emphasizes privacy focus and affordability
Plans announced for additional models targeting more complex problems
Key Takeaways: Sonus AI's entrance into the language model space with claimed top-tier performance metrics signals growing competition in the AI model landscape. While the benchmark scores are impressive, independent verification of these results will be important to validate the company's claims. The tiered model approach, from Mini to Pro, suggests a strategic focus on serving different market segments with varying computational and cost requirements. → Read more here.
Quick AI Headlines ⚡
Meta Expands AI Integration Across Platforms: Meta introduces AI characters on social media, aiming to increase engagement among its 3 billion users by rolling out AI tools that allow the creation of AI personas with bios and profile pictures on Facebook and Instagram. This initiative is part of Meta’s strategy to make its apps more entertaining as they compete with other tech companies for younger audiences. Connor Hayes, VP of product for generative AI, highlighted plans for AI-driven interaction and content creation. However, concerns over misinformation and content quality persist, stressing the need for safeguards.
OpenAI and Microsoft Outline AGI Economic Potential: OpenAI and Microsoft reveal AGI profit goals, designating artificial general intelligence (AGI) as technology capable of generating $100 billion in profits. This agreement from 2023 underscores their vision of AGI as systems surpassing human intelligence in economically significant tasks. While OpenAI strives for this milestone, projecting potential revenues by 2029, challenges remain given anticipated interim losses. The definition highlights AGI's potential transformative impact on the tech industry landscape.
Unitree B2-W Receives Exciting Upgrades: Unitree Robotics announces that one year after mass production began, their B2-W Industrial Wheel has been enhanced with advanced capabilities, including parkour skills, showcasing new industrial applications and emphasizing safe and friendly robot usage.
New York Enacts AI Monitoring Law for Government: New York state government to monitor AI use under new law, mandating state agencies to review and report their AI software usage publicly. Signed by Governor Kathy Hochul, the law prohibits AI in decisions about unemployment benefits and child care unless overseen by humans. This initiative, supported by State Sen. Kristen Gonzalez, aims to implement safeguards on AI's role in state operations, addressing concerns about automation impacts on state workers.
QVQ Revolutionizes AI Multimodal Reasoning: Qwen unveils QVQ model, advancing visual understanding and problem-solving capabilities. Built on Qwen2-VL-72B, QVQ scores 70.3 on MMMU benchmarks, outperforming predecessors in math-related tasks, and demonstrating powerful visual reasoning. Despite its achievements, the model faces challenges like language mixing and potential recursive reasoning. This development marks a step towards creating intelligent models capable of deep scientific exploration and complex problem-solving.
Google CEO Prepares for Critical Year Amid Intense Scrutiny: Google CEO Pichai addresses employees emphasizing the high stakes for 2025, as the company faces increasing competition and regulatory challenges. During a recent strategy meeting, Pichai highlighted the importance of accelerating efforts in AI advancements, particularly through the Gemini app, which aims to reach half a billion users. With ongoing regulatory scrutiny, including antitrust cases, Google plans to focus on efficiency and innovation while navigating a rapidly evolving tech landscape.
Nvidia Shifts Focus to Robotics Amidst AI Chip Rivalry: Nvidia plans robotics expansion by launching Jetson Thor for humanoid robots next year, aiming to become a leader as robotics demand rises. With competitors like AMD and tech giants entering the AI chip market, Nvidia invests in "physical AI" to support robotics companies, including Figure AI. Despite a smaller revenue share from robotics, Nvidia views this sector as pivotal amidst its data center dominance.
AI Job Market Booms with Focus on Leadership: Fast Company highlights AI's impact on jobs, revealing a threefold increase in AI-related leadership roles since 2022. With over 40% of businesses adopting AI, demand for senior leaders in AI has surged, with C-suite roles up 428%. This trend underscores the growing need for AI expertise in executive positions, vital for shaping strategies and driving innovation across industries. The competitive landscape requires companies to enhance recruitment and training to sustain AI implementation.
AI Unveils Secrets of Ancient Texts, Could Rewrite History: Nature explores AI’s role in unlocking ancient texts, highlighting projects like the Vesuvius Challenge, which used AI to read Greek papyrus scrolls carbonized in Mount Vesuvius's eruption. This breakthrough reveals texts unreadable for 2,000 years, reshaping historical research. By employing neural networks, researchers can decode damaged texts, explore vast archives, and potentially uncover a wealth of new historical data from sites like Herculaneum, offering a revolutionary new lens on ancient civilizations.
Big Tech Spends $125 Billion on AI Data Centers: Quartz reports on Big Tech's AI investments, highlighting a combined $125 billion spend by Microsoft, Meta, Google, and Amazon in 2024. Microsoft led with $40 billion, focusing heavily on GPUs and data center chips. Google and Amazon directed more resources towards AI model training, while Microsoft and Meta balanced expenditures between training and inferencing. This investment surge reflects the escalating demand for AI capabilities and data processing power.
Open-Source LLMs Match Proprietary Model Performance: Rollins College team demonstrates how open-source language models are achieving competitive results against closed-source leaders through innovative optimization techniques and collaborative development.
OpenAI Misses Deadline for Creator Opt-Out Tool: TechCrunch reports that OpenAI has not launched its promised Media Manager tool, which was to allow creators to control the use of their works in AI training. The absence of this tool leaves OpenAI facing multiple IP-related lawsuits and criticism for inadequate opt-out mechanisms. Despite claims of ongoing development, there's no clear timeline for its release, raising questions about the company's commitment to addressing content creators' concerns and legal challenges related to AI content training.
Samsung Expands into Robotics with Rainbow Robotics Investment: Samsung becomes largest shareholder in Rainbow Robotics, increasing its stake to 35% to advance robotic technology development. This strategic acquisition includes plans for Samsung to integrate Rainbow Robotics as a subsidiary. With this collaboration, Samsung aims to enhance its robotics capabilities, focusing on humanoid and autonomous robots, thereby solidifying its leadership in the global robotics market.
AI Robots Are Entering Public Spaces with Mixed Outcomes: The Wall Street Journal reports on the growing presence of AI robots in diverse public sectors. Enhanced by technologies like ChatGPT, these robots are expanding beyond factories to roles in retail, museums, and restaurants. Despite increased investment, with $12.8 billion raised in 2024, robots face challenges in performing tasks easily managed by humans, such as interacting in complex environments. As robotics progress, experts anticipate that generative AI will enhance capabilities, helping robots handle unforeseen tasks and improve interaction dynamics.
Alibaba Drastically Reduces AI Model Prices: Alibaba Cloud announces price cuts, lowering the cost of its visual language model, Qwen-VL, by up to 85% as competition in China's AI sector intensifies. This move underscores a strategic push to dominate the enterprise segment with affordable AI solutions. The price reduction follows previous discounts and highlights Alibaba's commitment to expanding its AI capabilities amidst growing interest in generative AI technologies from Chinese technology giants.
AI Orchestration Set to Transform Businesses in 2025: VentureBeat foresees 2025 as a pivotal year for AI orchestration, where integration and management of AI applications will streamline workflows across enterprises. Companies like AWS, Palantir, and Deloitte highlight the necessity for frameworks that effectively manage AI agents, with new platforms emerging to enhance agentic productivity. While better AI systems promise increased efficiency, the challenge remains in changing user behavior to fully realize AI's potential within corporate environments.