This week’s roundup for April 3, 2023 includes the latest news and information from the fields of data science, machine learning, and artificial intelligence. OpenAI, Microsoft, Meta, and Bloomberg are just a few of the major competitors in the market this week, and their new breakthroughs and innovations are the center of our attention.
Here we discuss the latest developments in natural language processing (NLP) models, new developer tools for AI, ethical AI practices, and other subjects.
OpenAI recently announced GPT-4, the next generation of the GPT family of Large Language Models (LLMs). GPT-4 can accept both text and image input and outperforms state-of-the-art systems on several natural language processing (NLP) benchmarks. The model also scored in the 90th percentile on the simulated bar exam.
OpenAI also unveiled an approach to AI safety by improving the ability of AI systems to learn from human feedback and assist humans in evaluating AI. The goal is to build a well-tuned AI system that helps solve all other coordination problems.
Microsoft provides an open source Semantic Kernel (SK). It is a lightweight SDK that enables integration of Large Language Models (LLM) with traditional programs, leveraging prompt templates, vectorized memory, intelligent planning, and other features.
Also, Microsoft researchers introduce TaskMatrix.AI. A new AI ecosystem that connects Foundation models with millions of APIs to complete tasks. The concept involves integrating foundational models with millions of existing models and system APIs, resulting in super AI that can perform a wide variety of digital and physical tasks. AI models and systems are currently designed to effectively address specific domains, but the diversity of their implementations and operating mechanisms can make them difficult for underlying models to access. . This new ecosystem aims to overcome these obstacles by providing a unified framework for connecting these AI models and systems.
Meta just released the Segment Anything Model. This new model allows AI to extract objects in images and videos with a single click. It uses advanced computer vision technology to enable computers to analyze and understand visual information such as images and videos. This is similar to how humans perceive and interpret what they see.
Researchers at Bloomberg and John Hopkins University are training BloombergGPT, a language model with 50 billion parameters useful for operations in various financial sectors. Rather than creating small or general purpose LLMs based solely on domain-specific data, they take a hybrid approach. We use standard his LLM criteria, open financial benchmarks, and Bloomberg’s proprietary benchmarks to evaluate our models and ensure they perform as expected.