The latest Deeper Insights blogs

The Costs and Complexities of Training Large Language Models

Explore the costs, complexities, and solutions in training large language models, crucial for advancing AI capabilities and managing associated challenges. [Read more]

‍Enhancing AI Responses with FLARE: A Game-Changer in Language Models

Discover how FLARE enhances AI by refining language model responses through targeted document retrieval, reducing inaccuracies, and improving reliability.  [Read more]

Custom LLM Development or Pre-built Solutions: What’s Right for Your Business?

Businesses face a choice: use pre-built, customise existing, or create new LLMs. Balancing cost, expertise, and benefits is crucial for optimal efficiency and innovation. [Read more]

How Synthetic Data is Reshaping Orthopaedics

Synthetic data revolutionises orthopaedics by enhancing surgical precision, AI training, and data privacy while overcoming data scarcity. [Read more]

Building Trustworthy AI through effective AI Governance

Discover how effective AI governance ensures responsible innovation, compliance with ethical standards, and manages AI-related risks for long-term trust and sustainability. [Read more]

Featured GenAI news

Nestlé Builds NesGPT: Safeguarding Internal Operations - WorkLife.news

Nestlé has developed NesGPT, a cutting-edge internal AI tool designed to transform its business operations. By addressing the issue of shadow AI, it ensures secure and compliant AI use within the company. This generative AI enhances content creation, customer engagement, and decision-making processes, bringing unprecedented efficiency and innovation. [Read more]

Making Sense of AI Terminology - TheVerge.com

Complex AI terminology is broken down to help readers understand the basics of artificial intelligence. Key concepts are explained in an easy-to-grasp manner, making them accessible to everyone. From machine learning to neural networks, this guide covers the essential terms you need to know to keep up with the rapidly evolving AI landscape. Perfect for anyone looking to get a clearer picture of AI technology. [Read more]

Responsible AI Implementation for CIOs - CIO.com

CIOs play a critical role in integrating generative AI responsibly within their organisations. To ensure ethical and effective AI deployment, CIOs should focus on enhancing data governance, implementing robust AI oversight, and fostering a culture of continuous learning. These steps help mitigate risks, streamline AI operations, and drive innovation, positioning companies for sustainable success in the rapidly evolving AI landscape. [Read more]

GenAI Enhances Creativity but Limits Diversity - Venturebeat.com

A recent study finds that while generative AI enhances individual creativity, it diminishes collective diversity. The research shows that AI-generated content improves personal creative output but may lead to homogenised ideas in group settings. Organisations are encouraged to balance AI use with initiatives to preserve diverse perspectives, ensuring innovation without compromising on creative diversity. [Read more]

GenAI news snapshots - Industry report

• The Simulation introduces Showrunner, a platform enabling users to create AI-generated TV shows. It supports various styles from anime to Pixar and South Park. Described as the "Netflix of AI", it offers both viewing and creation functionalities, highlighting a new era of interactive entertainment. [Read more]
• AI Transforms Climate Forecasting. Google's NeuralGCM model combines AI with traditional atmospheric physics, offering faster and more accurate forecasts. It successfully tracks climate trends and extreme weather events like cyclones. [Read more]
• AI boyfriends are transforming digital companionship with lifelike interactions that offer emotional support and romantic engagement. While expanding AI applications in business, they also raise concerns about social isolation and ethical implications. [Read more]
• OpenAI's GPT-4o mini is a new, affordable AI model that excels in text and multimodal reasoning. Offering exceptional accuracy and efficiency, it's perfect for tasks like customer support, coding, and data extraction. [Read more]
• Anthropic enhances its AI chatbot, Claude, with a new "tool use" feature, allowing users to create customised AI solutions like email assistants or shopping bots by connecting to external APIs. [Read more]
• GPs increase cancer detection rates by 8%. This advanced approach aims to improve early diagnosis, ensuring better patient outcomes and more effective treatments. [Read more]
• Toys ‘R’ Us utilised OpenAI's Sora, to create a unique AI-generated brand film. This marks one of the first commercial applications of the Sora technology, primarily showcased on the company’s website and at the Cannes Lions Festival. [Read more]
• Apple announced its integration of ChatGPT into iOS, iPadOS, and macOS. This allows users to access ChatGPT’s capabilities directly within Apple’s systemwide Writing Tools and through Siri. The integration will be available later this year. [Read more]

GenAI in Robotics

• LeRobot, hosted on Hugging Face, aims to lower the entry barrier to robotics by providing state-of-the-art models, datasets, and tools for real-world applications, focusing on imitation and reinforcement learning. [Read more]

• OpenAI is reviving its robotics research group after a three-year hiatus. The renewed focus comes as robotics companies increasingly integrate ChatGPT and LLMs. [Read more]

GenAI tools: LLM models

• Meta Chameleon is now publicly available! It offers a GPU-based standalone inference code for running model checkpoints and a browser-based tool for viewing multimodal inputs and outputs. It also includes evaluation prompts for both mixed-modal and text-only assessments. [Read more]
• Together MoA is a new Mixture of Agents (MoA) strategy which combines multiple open-source LLMs to set new benchmarks in AI performance. Its performance surpasses GPT-4o, highlighting the potential of collective intelligence in advancing LLM capabilities. [Read more]
• Qwen2 features models across five sizes, ranging from 0.5B to 72B parameters, now available on Hugging Face. Enhanced for performance in 29 languages, coding, and mathematics, it excels in various benchmarks and supports context lengths up to 128K tokens. [Read more]
• Mistral AI introduces model customisation on la Plateforme, enabling users to fine-tune and deploy tailored Mistral models with ease. This new service reduces the expertise and cost required for large-scale generative AI customisation, offering managed fine-tuning services and the flexibility to use your own infrastructure. [Read more]

GenAI tools: Visual models

• Stable Diffusion 3 Medium is the latest text-to-image model from Stability AI, designed for easy operation on consumer PCs, laptops, and enterprise GPUs. This model is accessible under a non-commercial open license or a low-cost Creator License, with a separate arrangement needed for large-scale commercial use. [Read more]
• Dragonfly model introduces an innovative vision-language architecture that utilizes a multi-resolution zoom-and-select feature for enhanced multi-modal reasoning. This model, along with its variants Llama-3-8b-Dragonfly-v1 and Llama-3-8b-Dragonfly-Med-v1, excels in commonsense visual QA and image captioning. [Read more]
• Mora is an open-source, innovative multi-agent framework designed to enhance generalist video generation, similar to OpenAI's Sora. Though comparable to Sora in video duration capabilities, Mora is still developing in areas like resolution and motion smoothness. [Read more]
• Gen-3 Alpha, by Runway, is a significant upgrade in video generation technology featuring high-fidelity, controllable outputs. It's designed for a broad spectrum of applications including Text to Video and Image to Video. [Read more]
• How to Fine-tune PaliGemma for Object Detection Tasks. This guide covers fine-tuning PaliGemma to detect fractures in X-ray images using a dataset from Roboflow Universe. The tutorial utilizes the smallest PaliGemma version (3B-224) to minimize GPU memory usage, making it suitable for Google Colab with an NVIDIA T4 GPU. [Read more]

GenAI tools: Everything else models

• Falcon 2 is a new suite of smaller, performance-enhanced, multimodal models designed for the open-source community. It aims to facilitate cheaper inference and broader application development. [Read more]
• Nomic Embed Vision includes high-quality vision models that share a unified latent space with its text models. The vision model, with its efficient 92M parameter design, is ideal for high-volume applications. Nomic also open-sourced the training code. [Read more]
• Claude 3.5 Sonnet is a notable upgrade from Anthropic, designed to surpass previous models in speed, intelligence, and cost-efficiency. It offers competitive pricing and enhanced accessibility, particularly for Pro and Team subscribers seeking higher usage limits. [Read more]

‍