Good Morning! Microsoft has unveiled exciting new AI capabilities coming soon to Copilot, including GPT-4 integration, better image generation, and new search and code features. The Onyx programming language offers developers a modern, performant language with C-compatibility, WebAssembly support, and a focus on pragmatic, efficient coding. Researchers at DeepMind have developed AI agents that demonstrate social learning abilities, acquiring knowledge faster by observing expert agents, much like humans learn skills through cultural transmission.

— Forrest Knight & Meghanadh Vasireddy

Microsoft Unveils Upcoming Copilot Features

Microsoft recently announced upcoming new features coming soon to its AI assistant, Microsoft Copilot. Copilot brings together several Microsoft AI products like Bing Chat, Microsoft 365, Edge, and Windows to provide a unified AI experience.  

The biggest update is that Copilot will soon integrate GPT-4 Turbo, OpenAI's latest AI model allowing it to handle more complex tasks. A few testers already have access, with a full rollout expected in weeks. Another new addition is the DALL-E 3 model for higher quality AI image generation from text prompts. This will be great for graphic designers, marketers, and other creative professionals.

Edge will gain the ability to select text on a webpage and have Copilot rewrite it for you. Helpful for simplifying complex passages or summarizing key information.

Bing gains Deep Search, using GPT-4 to expand search queries into more detailed descriptions. This helps Bing return more relevant results. Also coming is improved image search using GPT-4's computer vision capabilities. A new code interpreter feature is in development, allowing Copilot to handle calculations, data analysis, coding, math and more. Microsoft is currently gathering feedback before release.

"This year will be remembered as the moment that we...began to harness the power of AI in our daily lives," said Microsoft's Yusuf Mehdi. "The last 10 months reflect years of AI research, partnerships, and innovations coming together."

Read More Here

Onyx: A Modern, Performant Programming Language

The Onyx programming language is a modern, data-oriented, and expressive language that offers a unique blend of features designed to enhance the programming experience. It uses a modernized C-like syntax, similar to Jai or Odin, making it familiar to many developers.

Onyx is strictly type-checked, but its type-inference systems usually allow you to omit types, making the code cleaner and easier to read. It also boasts fast compilation times, with the compiler written entirely in C. This language is designed to be pragmatic, aiming to facilitate getting things done efficiently.

One of the standout features of Onyx is its built-in support for linking to native C-libraries, which can be dynamically loaded at runtime. This makes it a powerful tool for developers who need to interface with existing C codebases.

Onyx is also designed with performance in mind, and it compiles to high-performance native machine code. It's suitable for both application and system programming, making it a versatile choice for a wide range of projects.

Moreover, Onyx leverages WebAssembly (WASM) for seamless cross-platform support. This means that programs written in Onyx can be run on any platform that supports WASM, providing a high degree of portability.

The language also emphasizes code readability and conciseness, enabling developers to express complex concepts in a clean and straightforward manner. It also supports asynchronous and concurrent programming, allowing developers to write highly performant code.

Read More Here

Advances in Social Learning for AI

Researchers at DeepMind have developed an artificial intelligence system that demonstrates an ability to learn socially, much like humans do. Their work, published in Nature Communications, marks an important step towards more versatile and efficient AI.

The AI agents inhabit a 3D simulated world with challenging terrain, obstacles, and goals to reach. Without any instructions, they use trial-and-error and reinforcement learning to figure out routes. But when the researchers introduced "expert" agents that already knew optimal paths, the other AI agents learned much faster by observing and mimicking them.

This cultural transmission enabled the AI systems to navigate new environments better after just a little exposure from the experts. And they retained the knowledge even when the experts were no longer present. The researchers say this is similar to how humans learn skills quickly from observing others, instead of through repeated individual trial-and-error.

Current AI systems like ChatGPT amass knowledge over massive datasets, which is data-intensive and computationally expensive. In contrast, human learning is extremely sample-efficient, picking up new skills with just a few examples. Enabling AI to learn socially and culturally could make future systems more versatile, energy-efficient, and scalable.

The DeepMind team notes their work is an initial step with limitations. The simulated world is simple compared to the real world. And the social learning dynamics between multiple AI agents remains to be explored. However, their research demonstrates the promise of such approaches to develop AI that learns more like humans do.

Read More Here

MongoDB Energizes Generative AI Capabilities

MongoDB, the popular document-based database platform, recently announced new products to help companies more easily build and scale generative AI into their applications.

With MongoDB Atlas Vector Search now generally available, developers can enable AI features like semantic search and image comparison using MongoDB's flexible data model. Requests can combine vector search queries with other data like locations, dates, text search, and more.

For example, a real estate application could find listings matching a photo and user preferences on location, school districts, and nearby parks. The rich unified queries produce more relevant, customized results.

To empower these complex generative workloads, MongoDB also launched Atlas Search Nodes - dedicated infrastructure isolated from the main operational database. Companies can optimize cost and performance for AI search separately from transactional or analytical workloads.

A retailer could provision search nodes to handle increased chatbot queries during peak holiday shopping, without affecting its core e-commerce system. Atlas Search Nodes reduce vector search query latency by up to 60%.

"With the general availability of MongoDB Atlas Vector Search and MongoDB Atlas Search Nodes, we're making it even easier for customers to use a unified, fully managed developer data platform to seamlessly build, deploy, and scale modern applications and provide end users with the types of personalized, AI-powered experiences that save them time and keep them engaged," said Sahir Azam, MongoDB's Chief Product Officer.

Read More Here

Internet Spotlight

A Programmer Presents a Mind-Blowing Multi-Window Adaptive Experience

“Bjorn Staal, an independent Artist, Programmer, and Researcher, has recently blown the minds of thousands of people by unveiling a magical setup that allows one to "synchronize" a 3D scene across multiple browser windows. Powered by Three.js and localStorage, the project showcases a distinctive approach to developing and overseeing a 3D scene across various browser windows. According to the creator, this outstanding setup was designed primarily for developers keen on exploring advanced web graphics and window management techniques.”

- 80LV

Was this forwarded to you? Sign Up Here