The latest Deeper Insights blogs

From Bias to Balance: The Crucial Role of Diversity in AI

Diversity in AI is crucial to counter biases and ensure technology equitably serves all. By integrating varied perspectives in data and development, we can create inclusive AI solutions that reflect the diversity of our global community and address societal inequalities effectively. [Read more]

Revolutionising Educational Environments Through AI

Artificial Intelligence is revolutionising education, enhancing personalised learning and outcomes. AI extends beyond automation, reshaping teaching and enriching knowledge acquisition for all students, including those with special needs. [Read more]

Robots That Care: Designing Robots to Enhance Human Interaction

There is tremendous transformative potential in emotionally intelligent robots. We have the technology to build AI infused robots today. Learn how adding emotional intelligence can enhance human-machine interactions across various sectors. [Read more]

Deepfake Reality: Promise vs Peril

Deepfakes blend reality and AI, presenting marvels and ethical dilemmas. They amuse and distort truth, raising concerns. Balancing their use is crucial for leveraging benefits while protecting reality. [Read more]

Featured GenAI news

OpenAI is being sued by Elon Musk - CNBC.com

Elon Musk has filed a lawsuit against Microsoft-supported OpenAI, including CEO Sam Altman, accusing them of straying from their original goal of creating AI for the widespread benefit of humanity. In response, OpenAI has labelled Musk's lawsuit as 'frivolous' and 'incoherent' in their legal filings. This lawsuit brings two leading figures in technology into conflict amid intense excitement about AI's potential future. [Read more]

U.S. Government Sues Apple Over Market Monopoly - BBC.co.uk

The U.S. Department of Justice has initiated a significant lawsuit against Apple, alleging that the company has monopolized the smartphone market through unfair practices that deter competition. According to the lawsuit, Apple has manipulated its app store and hardware to hinder rival products and services, thereby maintaining its market dominance. This legal challenge, supported by the attorneys general of 16 states, could potentially compel Apple to change its business practices and is part of a broader trend of regulatory scrutiny against major tech companies. [Read more]

Generative AI Could Speed Up 80% of Jobs - Google.com

Google's latest report, co-authored by Andrew McAfee, explores the economic implications of generative AI. Defined as a "general-purpose technology," it has the potential to significantly enhance productivity by accelerating task completion by 10% across 80% of U.S. jobs. The research, backed by experts at Google, underscores the capacity of generative AI to drive economic growth by boosting efficiency in a wide range of industries, suggesting a substantial impact on the future workforce and economic landscape. [Read more]

DARPA Advances Military Tech with AI - DARPA.mil

DARPA's RACER program has successfully tested autonomous movement on a new, larger fleet tank, marking a significant advancement in the adaptability and capability of its underlying algorithms. This expansion to the RACER Heavy Platform vehicles, which are notably larger at 12 tons, demonstrates significant progress in scaling up the technology for broader applications. [Read more]

GenAI news snapshots - Industry report

• The U.S. has passed a bill that could ban TikTok if its Chinese parent company, ByteDance, does not sell its stake in the app. The legislation, aimed at ending Chinese ownership of TikTok, has moved through the House with bipartisan support, though it's uncertain if a sale would be approved by China or who might purchase TikTok. This potential move raises questions about similar actions in the UK. [Read more]

• Amazon is allocating $1 billion to boost startups merging AI with robotics, aiming to enhance its logistic efficiencies. This effort is part of a broader strategy to automate warehouse operations and improve delivery systems without fully replacing human jobs, even amid a tech downturn. [Read more]
  

• Apple launches on-device AI, OpenELM, an open-source suite of large language models tailored for on-device execution, enhancing privacy and performance. Available through Hugging Face. With its focus on on-device AI, Apple promotes a shift toward more private, efficient AI solutions, marking a significant step in distributed AI technology. [Read more]
  

• MagicLab's MagicBot is a humanoid robot capable of performing tasks like toasting marshmallows, folding clothes, and dancing, showcasing high dexterity and adaptability. Developed with advanced servo actuators and pressure sensors, it mimics 70% of human hand gestures, promising applications across various sectors. [Read more]
  

• OpenAI collaborates with Figure AI to develop advanced AI models for humanoid robots, aiming to integrate them into everyday life and work, targeting sectors like manufacturing and logistics to address labour shortages. Figure AI, valued at $2.6 billion, raises $675 million in Series B funding. [Read more]
  

• Covariant is launching RFM-1, a platform aimed at giving robots the ability to process and reason from language, akin to a "ChatGPT for robots." This initiative, developed by the UC Berkeley spinout, aims to enhance robots' decision-making by providing a human-like understanding of language and the physical world, promising to revolutionize tasks in manufacturing, logistics, and beyond. [Read more]
  

• Physical Intelligence has secured $70 million in seed funding to create AI-powered robots versatile enough for any use case. This initiative, spearheaded by both industry and academia leaders blends advanced language model technologies with innovative machine control techniques. Targeting a wide array of sectors, including manufacturing, logistics, and healthcare, the company aims to redefine robot functionality with its universally applicable software, compatible with various robotic hardware. [Read more]
  

• Apple makes a commitment to AI, Apple's CEO Tim Cook has announced the company's pledge to pioneering developments in GenAI this year. This declaration was made at Apple's annual shareholders meeting, coinciding with reports of the company abandoning its ambitious electric vehicle project, a venture spanning over a decade and worth billions. [Read more]
  

• Adobe Research is innovating audio creation and editing with Project Music GenAI Control, a generative AI tool allowing creators to generate and finely edit music from text prompts. This tool offers a new level of control for crafting music tailored to specific moods, tones, and lengths, integrating seamlessly into creative workflows. [Read more]
  

• Midjourney introduces a "consistent characters" feature, enabling the same character to maintain their unique design across various images and styles. Detailed usage instructions are available on Midjourney's Discord channel, optimized for generated characters rather than real people. [Read more]
  

• Anthropic introduces the Claude 3 family: Claude 3 Haiku, Sonnet, and Opus, each designed for different levels of intelligence, speed, and cost efficiency. Opus stands out for its high intelligence, Sonnet balances speed and smarts, and Haiku is fast and cost-effective. They offer better vision capabilities, lower refusal rates, enhanced accuracy, and can handle longer contexts. [Read more]
  

• Adobe Acrobat's new AI Assistant, in beta, allows users to quickly interact with PDFs for answers and summaries, enhancing document productivity. This feature, for English-speaking paid and trial users, integrates advanced language and generative AI technologies, focusing on ethical principles like accountability and transparency, aimed at improving content engagement and creation securely. [Read more]
  

• OpenAI introduces ChatGPT's "Read Aloud" feature, enabling it to audibly deliver responses in multiple languages and voice options on web and mobile platforms. This function, supporting GPT-4 and GPT-3.5, enhances user interaction by reading answers out loud, with options to play, pause, and rewind, facilitating use while on the move. [Read more]
  

• Google users can now fine-tune AI responses on Gemini by highlighting text and selecting options like Regenerate, Shorter, Longer, and Remove. This new feature aims to refine outputs for a closer match to user expectations, enhancing creative control over content. [Read more]

GenAI tools: LLM models

• Rime's Mist is a groundbreaking conversational voice synthesis tool offering ultra-low latency and realistic speech patterns, including natural-sounding filler words. It boasts a wide range of accents and voice types from a diverse demographic dataset, aiming to enhance conversational AI with hyperrealistic, nuanced voices suitable for various applications. [Read more]
  

• Inflection-2.5 enhances their AI, Pi, balancing IQ and EQ to rival GPT-4 with less compute. It excels in coding, and mathematics, and includes real-time web search. Serving over a million daily users, Inflection-2.5 blends intelligence, safety, and personality, positioning Pi as a leading personal AI. [Read more]
  

• "C4AI Command-R v01" is a 35 billion parameter AI model by CohereForAI, designed for reasoning, summarisation, and Q&A across 10 languages. It features RAG for advanced generation and is aimed at facilitating community-based AI research by sharing model weights. This model enhances conversational and grounded generation capabilities, contributing to AI advancements. [Read more]
  

• Training-Free Long-Context Scaling of LLMs ChunkLlama uses dual chunk attention to extend the context capabilities of language models like Llama2, allowing them to process contexts over eight times longer than originally designed, without additional training. This innovation enhances performance on long-context tasks, offering a new direction for efficient language model utilisation. [Read more]
  

• The "text-clustering" project on GitHub is a toolkit for embedding, clustering, and labeling text datasets semantically, offering a simple, adaptable approach for text data analysis. It enables users to perform these tasks quickly and efficiently on standard laptops, using well-known libraries like scikit-learn. This project is ideal for researchers and practitioners needing to organize and understand large text corpora with minimal setup.[Read more]
  

• Answer.AI is designed for collaborating with other experts, introduced a method to train a 70 billion parameter model on desktops with 24GB GPUs, making advanced AI model development more accessible to smaller entities. Utilizing FSDP and QLoRA, this initiative aims to democratize AI research, allowing for the creation of personalized models with fewer resources. [Read more]

GenAI tools: Visual models

• Genie is an AI that generates playable, interactive environments from images. Leveraging Internet videos for training without specific action labels it enables dynamic control across various generated worlds. This technology fosters creative virtual space generation and aids in the development of AI agents by providing a limitless array of training environments, heralding a new era in generative AI and interactive experiences. [Read more]
• The All-Seeing Project V2 introduces the ASMv2 model for comprehending object relations in images, integrating text generation, object localisation, and relation understanding. It includes a high-quality dataset and a new benchmark, CRPE, for evaluating models. [Read more]
• Sakana.ai announces the launch of two advanced Japanese foundation models, Large Language Model (EvoLLM-JP) and Vision-Language Model (EvoVLM-JP), available on Hugging Face and GitHub, and hints at the upcoming release of Image Generation Model (EvoSDXL-JP). [Read more]

GenAI tools: Everything else models

• "Moondream" is a compact vision-language model that stands out for its ability to understand and generate text based on images. It's built upon models like SigLIP and Phi 1.5, with 1.86B parameters, and excels in benchmarks like VQAv2, GQA, and TextVQA. It's designed for generating descriptive texts from images, highlighting its capability in linking visual perception with language processing. [Read more]
  
• The 3D Diffusion Policy (DP3) is a method for teaching robots skills through visual imitation, using 3D visual representations from point clouds. It achieves high success in both simulated and real tasks, demonstrating strong generalization capabilities. [Read more]

‍