The latest Deeper Insights blogs

How GenAI is Revolutionising the Business: A Guide for Today’s Leaders

Studies show that GenAI can boost productivity by up to 60% and contribute trillions to the global economy. While risks exist, they're manageable with expertise, making GenAI adoption a strategic necessity for modern businesses. [Read more]

The Human Blueprint for Teaching Machines to Recognise Faces

Explore the intricate world of facial recognition through the lenses of neuroscience, psychology, and artificial intelligence. Discover how our brains are wired to recognise faces, the challenges posed by conditions like prosopagnosia. [Read more]

A New Era of Healthcare: Navigating Synthetic Data

Synthetic data is emerging as a transformative force with applications ranging from medical research to patient care. This overview explores how synthetic data is revolutionising the study of stigmatised illnesses and its role in the COVID-19 pandemic. It also reviews the technology's advantages, such as enabling cross-institutional research, and its limitations like data representation and privacy concerns. With insights into ongoing initiatives, this piece offers a well-rounded look at the current and future impact of synthetic data in healthcare. [Read more]

Featured GenAI news

Meta Unveils Cutting-Edge Artificial Intelligence Chatbots

Meta introduces a new artificial intelligence chatbot, showcasing enhanced conversational capabilities. This launch signifies Meta's continued efforts in pushing the boundaries of AI and enhancing user engagement on its platforms. Through this innovative chatbot, Meta aims to provide a more natural and intuitive user experience, marking a significant advancement in AI-driven communication technology. The company's investment in artificial intelligence underscores its commitment to fostering a more interactive and user-centric digital ecosystem. [Read more]

Google Ads Upgrades Suite with AI  

Google Ads is enhancing its digital advertising suite by integrating AI-driven auto-generated assets and a dialogue-based feature designed to personalise and fine-tune search ads for marketers. [Read more]

Microsoft Eyes Next-Gen Nuclear Reactors to Fuel Data Centers

Microsoft is planning to leverage advanced nuclear reactors to power its data centers and support its AI projects. To this end, the company has listed an opening for a principal program manager to oversee its nuclear energy initiatives. Given the high energy consumption of data centers, Microsoft aims to utilise nuclear technology as a clean energy source to meet its environmental objectives. [Read more]

OpenAI May Secure Funding at a Staggering $80-90 Billion Valuation

OpenAI reportedly in active talks to sell its shares, a move that could skyrocket the company's valuation from its current $29 billion to a staggering $80 billion. Additionally, employees have been given the green light to sell their existing shares in the company. In a statement made in late August, OpenAI anticipates reaching an impressive $1 billion in revenue this year. [Read more]

GenAI news snapshots - Industry report

• ChatGPT Elevates Multimodal Interaction with Hearing, Speaking, and Visual Capabilities. OpenAI enhances ChatGPT with multimodal abilities: audio input, text-to-speech, and image recognition, paving the way for more intuitive user interactions. These updates position ChatGPT as a more versatile tool in the evolving AI landscape. [Read more]
• Mistral AI releases its first expansive language model to the public at no cost. As AI communication tools gain traction, this initiative marks a significant stride towards fostering innovation and accessibility in artificial intelligence realms. [Read more]
• OpenAI Red Teaming Network - a team of experts aimed at enhancing AI model risk assessment and mitigation strategies. As AI technologies, especially generative ones, become widespread, red teaming becomes pivotal in identifying and addressing biases and safety issues. [Read more]
• Llama 3 is rumored to be on par with GPT-4 in terms of performance, and it will continue to be accessible under the Llama license, as shared by OpenAI engineer Jason Wei during a social event organized by Meta. [Read more]
• Elon Musk cautions senators about AI risks, while Bill Gates suggests AI's potential to address global hunger during a closed-door Senate session on the topic of artificial intelligence. [Read more]
• Adobe's Firefly generative AI models are now officially available in Creative Cloud, Adobe Express, and Adobe Experience Cloud after 176 days in beta. Users can enjoy features like generative fill and expand in Photoshop without the need for beta installations. [Read more]
• Elevate productivity with advanced security, unlimited GPT-4 access, longer input processing, and customization, enhancing your work AI experience while safeguarding company data using ChatGPT Enterprise. [Read more]
• OpenAI releases a guide for teachers using ChatGPT covers prompts, AI functionality, limitations, detectors, and addressing bias in the classroom. [Read more]
• Generate Biomedicines: A groundbreaking therapeutics company merging AI, biology, and medicine to revolutionize drug development. Their "Generate Platform" has been advocate to have limitless technical potential. [Read more]
• Google's Bard AI chatbot goes beyond web searches, now scanning Gmail, Docs, and Drive to retrieve information efficiently. This integration, called extensions, streamlines tasks like summarizing emails and highlighting key document points, with more use cases on the horizon, albeit currently in English only. [Read more]
• MidJourney's introduces "Vary (Region)", enabling users to modify specific parts of images using text prompts. However, competition with Stable Diffusion reveals limitations in accuracy and user experience. [Read more]
• Consensus AI is a search engine powered by AI and GPT-4 that swiftly extracts evidence-based answers from scientific research. It democratises expert knowledge, offering reliable insights on various topics. With ad-free access, customizable searches, and easy-to-understand summaries, Consensus AI makes accessing verified scientific information faster and simpler for everyone. [Read more]
• Snapchat is venturing further into generative AI with its upcoming feature "Dreams," enabling users to generate imaginative AI images by uploading selfies, with potential collaborative options. This move reflects Snapchat's innovative AI-driven approach. [Read more]
• LoRA the Explorer is a community-developed space with LoRAs for SDXL. Pic a style and have fun! Check a video demo here. [Read more]
• Google's Search Generative Experience (SGE) introduces AI tools for term definitions, coding help, and content summaries. It lets users preview word meanings and code segments, and experimentally offers AI-generated content summaries during web browsing. [Read more]

GenAI tools: LLM models

• Falcon 180B - the powerhouse language model with 180 billion parameters, outperforming in reasoning, coding, proficiency, and knowledge tests. It currently leads Hugging Face's leaderboard for large language models and rivals even larger models like GPT-4 and PaLM 2 Large while being just half their size. Available for both research and commercial use. [Read more]

• Persimmon-8B - an open-source model that is part of Adept's journey towards creating a versatile AI agent for various computer tasks, reflecting their commitment to evolving beyond standalone language models. [Read more]

• TinyLlama - Pretraining a 1.1B Llama model on 3 trillion tokens is underway, targeting completion in just 90 days with 16 A100-40G GPUs. This model shares the same architecture and tokenizer as Llama 2, making it easily integrated into various open-source projects, all while being compact with 1.1B parameters for efficient use in memory-restricted applications. [Read more]
• Nougat - a Visual Transformer model. Nougat performs OCR on scientific documents, converting them into a markup language. This innovative approach enhances accessibility by making scientific knowledge machine-readable, closing the gap between human and machine understanding. [Read more]

• medGPT - Unlocking LLM Potential for Clinical Medicine: While LLMs possess impressive natural language skills, aligning them effectively for clinical applications is key. The 'expand-guess-refine' strategy with instruction-tuning and prompt techniques, proves a data-efficient solution. Early results shine, with a remarkable 70.63% score on a subset of USMLE questions. [Read more]
• OctoPack - enhance code language models using Git commit instructions. The authors of introduce CommitPack, a dataset from Git commits in various languages, and HumanEvalPack, a diverse benchmark. This approach shows improved performance, with models OctoCoder and OctoGeeX excelling in HumanEvalPack. [Read more]

• PlayHT1.0 - a text-to-speech model that creates remarkably lifelike and emotion-infused speech, including laughter. With its unique self-supervised approach, the model offers diverse voices and styles, even enabling voice cloning from just 30 seconds of audio. This advancement holds potential for applications across industries, revolutionizing content creation and voice production. [Read more]
• SeamlessM4T - an all-in-one multimodal translation model, pioneers speech-to-speech and speech-to-text translation with unprecedented language coverage and accuracy. It supports translation of nearly 100 input languages and 35 output languages. This breakthrough model simplifies communication across languages, delivering high-quality results and reducing reliance on multiple models. [Read more]

GenAI tools: Visual models

• DALL·E 3 - Elevating Text-to-Image Accuracy. This new model understands nuances better, translating your ideas into highly accurate images without the need for complex prompts. Say goodbye to prompt engineering with this leap forward in text-to-image generation. [Read more]
• FreeU -  a simple but effective method that enhances the quality of generated images and videos without additional training. It optimizes the U-Net architecture to improve generation quality, and it can be easily integrated into existing diffusion models with minimal code changes. [Read more]
• DatasetDM - introduces a data generation model utilizing diffusion models to create diverse synthetic images with perception annotations. The method efficiently decodes latent information, enabling the generation of large annotated datasets for training perception models across tasks such as segmentation and depth estimation, showcasing strong performance and flexibility. [Read more]
• Fooocus - merges Stable Diffusion and Midjourney designs to provide a user-friendly, open-source image generation software. With streamlined processes and automated enhancements, users can focus solely on prompts and images, unlocking new creative possibilities. The software's accessible installation and advanced features make it a noteworthy tool for AI-assisted artistic exploration. [Read more]
• Virtual Try - On with Diffusion Models and Warping - A novel technique for improving virtual try-on using diffusion models. By combining a warping module with the diffusion process, the method effectively enhances the quality of results by preserving clothing details and achieving superior realism. Experimental validation on the VITON-HD dataset confirms the method's efficacy, making it a valuable advancement for virtual try-on applications. [Read more]

GenAI tools: Everything else models

• Council: Open-Source AI Agent Platform for Controlled Generative Applications. Council is a Python-based open-source platform that facilitates the development and deployment of customisable generative AI applications through controlled agent behavior. The framework supports various LLMs, ecosystem connectivity, and scalable oversight. [Read more]
• DeepEval: Streamlined Testing for Language Model Pipelines. DeepEval simplifies offline evaluations of LLM pipelines, akin to a "Pytest for LLM”. This tool bridges the gap between conventional testing methods and the unique evaluation requirements of AI models, making iterative testing and deployment of LLMs efficient and practical. [Read more]
• Opendream: Streamlined UI for Advanced Image Editing. Opendream streamlines Stable Diffusion workflows with non-destructive editing, layering, and effortless extension integration. This interface empowers creative experimentation, facilitates workflow sharing, and simplifies the incorporation of new features, enhancing image manipulation processes. [Read more]
• Multi-Persona Collaboration Enhances LLM Performance. Solo Performance Prompting (SPP) is a method that enables Large Language Models (LLMs) to engage in multi-turn self-collaboration with various personas. This approach, aimed at complex tasks, harnesses cognitive synergy by combining strengths and knowledge from different personas. Through dynamic persona simulation, SPP improves problem-solving abilities, knowledge acquisition, and reasoning in LLMs, as demonstrated in tasks like Trivia Creative Writing, Codenames Collaborative, and Logic Grid Puzzle. [Read more]

‍