top of page

AI Insights

Your weekly newsletter

It was ‘Everything Everywhere all at Once’ moment for AI this week. Google, OpenAI, MidJourney, Microsoft, Stanford and others launched new AI products.


  • Stanford University researchers release Alpaca, a fine-tuned instruction-following language model to address the deficiencies of existing models. | Source

A team of researchers from Stanford University has released Alpaca, a fine-tuned instruction-following language model that is surprisingly small and easy to reproduce. Alpaca is designed to address some of the deficiencies of existing instruction-following models, such as generating false information, propagating social stereotypes, and producing toxic language. The researchers are releasing their training recipe and data, as well as an interactive demo, to enable the research community to better understand the behavior of Alpaca.

  • Google shares updates on its health AI research, including progress on its medical large language model and partnerships for AI-assisted ultrasound, tuberculosis screening, and cancer treatment planning. | Source

Google has shared updates on its health AI research, including progress on its medical large language model (LLM) and partnerships to bring AI-assisted ultrasound and tuberculosis screening to areas with limited access to healthcare. The company is also partnering with Mayo Clinic to explore how AI can support the process of planning for radiotherapy in cancer treatment.


  • OpenAI releases GPT-4, the latest and largest language model, and introduces OpenAI Evals for evaluating models. | Source

OpenAI has released GPT-4, the latest and largest language model developed by the company, with 6 trillion parameters. The model has the potential to benefit various applications, including language translation, chatbots, and content creation, but also comes with potential risks and challenges associated with the technology. OpenAI has also introduced OpenAI Evals, a software framework for evaluating models like GPT-4, and offers access to the model through the ChatGPT Plus subscription and API.

  • Google Cloud announces partnerships and new tools for the advancement of generative AI, including generative AI support in Vertex AI and Built with Google Cloud AI. | Source

Google Cloud has announced partnerships and new tools for the advancement of generative AI, aimed at accelerating the development of new and innovative AI capabilities. The tools include new generative AI support in Vertex AI, generative AI app builder, and Google Workspace, as well as new initiatives like Built with Google Cloud AI, which recognizes partners who have built applications using Google Cloud's AI capabilities.

  • Anthropic launches Claude, an AI assistant for conversational and text processing tasks | Source

Anthropic has launched Claude, an AI assistant designed to help with conversational and text processing tasks such as summarization, Q&A, search, coding, creative and collaborative writing, and more. Claude is based on Anthropic's research into training AI systems to be helpful, honest, and harmless. It is available through chat interface and API in the developer console, with two versions - Claude, a high-performance model, and Claude Instant, a lighter, less expensive, and much faster option. Anthropic's partners, including Notion, Quora, and DuckDuckGo, are already using Claude to power their online education, legal infrastructure, and search solutions.


  • PyTorch 2.0 Released with Faster Performance, Dynamic Shapes, and Distributed Support | Source

The PyTorch team has announced the release of PyTorch 2.0, which comes with faster performance, dynamic shapes, and distributed support. The new release offers the same user experience as before while fundamentally changing how PyTorch operates at the compiler level under the hood. PyTorch 2.0 includes a stable version of Accelerated Transformers and beta versions of other libraries, including TorchAudio, TorchVision, and TorchText. The release includes other prototype features and technologies across TensorParallel, DTensor, 2D parallel, TorchDynamo, AOTAutograd, PrimTorch, and TorchInductor. The PyTorch team encourages users to try out the new release and report any issues to improve it.

  • Midjourney Releases Latest Model, v5, with Improved Photorealistic Images and Accurate Human Hands | Source

Midjourney has released the latest version of its AI image generator, Midjourney v5, which boasts improved photorealistic images and accurate human hands. The new version has a much wider stylistic range, offering a wide diversity of outputs that are more responsive to user inputs. Users will get higher image quality and improved dynamic range, but will need to provide more detailed text prompts to achieve the best results. Although Midjourney v5 could extend the toolbox of artists and photographers, it is currently only available to paying customers.

  • Stripe partners with OpenAI to enable payments for ChatGPT Plus and DALL·E and enhance its platform with GPT-4 | Source

Stripe and OpenAI have announced a collaboration where Stripe will power payments for OpenAI's ChatGPT and DALL·E generative AI technologies. Stripe is also incorporating OpenAI's new natural language technology, GPT-4, into its products and services. The collaboration includes using Stripe Billing and Stripe Checkout to power OpenAI's subscription offering of ChatGPT Plus and offering pay-as-you-go options for purchasing DALL·E credits. In addition, OpenAI users can pay 40% faster on average with Stripe's autofilling saved payment details. Stripe has also announced GPT-powered Stripe Docs, allowing developers to pose natural language queries and get answers from GPT-4.


  • Microsoft launches Microsoft 365 Copilot, an AI-powered productivity tool to revolutionize the way people work. | Source

Microsoft has launched Microsoft 365 Copilot, an AI-powered productivity tool that aims to revolutionize the way people work by unlocking productivity and creating a new knowledge model for organizations. The tool is designed to learn new skills and will be integrated into all productivity apps, including Word, Excel, PowerPoint, Outlook, Teams, Viva, Power Platform, and more in the months ahead.


  • Microsoft releases Semantic Kernel, a lightweight SDK for integrating AI capabilities into apps | Source

Microsoft has released an open-source framework, Semantic Kernel (SK), designed to quickly integrate Large Language Model (LLM) AI "prompts" into applications, without needing to train or fine-tune a model from scratch. SK lets developers mix conventional programming languages with the latest in LLM AI "prompts" with prompt templating, chaining, and planning capabilities. SK supports models from OpenAI, including the newly released GPT-4, and Azure OpenAI Service. It is lightweight and extensible, allowing developers to connect with external data sources and services, and can be embedded in any kind of application. SK is available on GitHub with examples to get started quickly. LinkedIn Learning also offers a free course on Semantic Kernel.

bottom of page