. . AI: Google's New Small Vision Model and Anthropic Models in Google Cloud (3.20.24)

In today's edition, we explore Google's new 5B parameter vision-language model, as well as the news that Anthropic's models are now accessible through Google Cloud. We also highlight resources and tools relevant to developers, ranging from front-end to neural networks!

As always, these are the stories the AI community is talking about most. We hope you find them useful, literally and/or as analogies and new ways to think about the world!

-Marshall Kirkpatrick, Editor

Google's AI Team Launches ScreenAI

First impacted: AI researchers, UI/UX designers
Time to impact: Short

Google's AI team has launched ScreenAI, a vision-language model designed to interact with user interfaces and infographics. Google says that ScreenAI, with its 5 billion parameters, has achieved top results on tasks involving user interfaces and infographics, and has outperformed similar-sized models on Chart QA, DocVQA, and InfographicVQA. The team also announced they are releasing three new datasets for further research and evaluation of ScreenAI's capabilities. [ScreenAI: A visual language model for UI and visually-situated language understanding] Explore more of our coverage of: Google AI, User Interfaces, Infographics Analysis. Share this story by email

Anthropic's Models Now on Google Cloud

First impacted: AI solution developers in businesses, Quora's Poe app users
Time to impact: Short

Anthropic announced that their models Claude 3 Haiku and Claude 3 Sonnet, are now accessible on Google Cloud's Vertex AI platform, enabling businesses to develop and scale AI solutions with robust data privacy and security. Anthropic plans to release its most advanced model, Claude 3 Opus in the near future too. They also shared that Quora's Poe app has integrated Claude, leading to millions of daily message exchanges, though Opus is only available to paid Poe subscribers. [Claude 3 models on Vertex AI] Explore more of our coverage of: Anthropic AI, Google Cloud, AI Integration. Share this story by email

Puzzles for Triton Developers

First impacted: Programmers, AI developers
Time to impact: Short

Sasha Rush has shared the 7th set of interactive puzzles, this time for Triton, an open-source language for accelerators like GPUs. The puzzles, designed to teach Triton usage from basic to advanced levels and includes algorithms such as Flash Attention and Quantized neural networks, and can be run on a Triton interpreter. [GitHub - srush/Triton-Puzzles: Puzzles for learning Triton] Explore more of our coverage of: Open-Source Language, GPU Accelerators, Neural Networks. Share this story by email

Maestro Python Script Launched for Developers to do Task Breakdowns

First impacted: Software Developers, AI Engineers
Time to impact: Short

Maestro, which is a python script, has been launched using the Anthropic API and two AI models, Opus and Haiku. According to the developers, Maestro breaks down tasks into smaller components, executes them with the Haiku model, refines the results with the Opus model, and generates a detailed interaction log, all customizable to the user's needs. [GitHub - Doriandarko/maestro: A framework for Claude Opus to intelligently orchestrate subagents.] Explore more of our coverage of: Python Script, AI Models, Task Automation. Share this story by email

Glowby Launches Draw-to-Code AI Assistant

First impacted: Software Developers, App Designers
Time to impact: Short

Jacob Ilin has launched Glowby, a GenAI assistant for software development, which he says allows users to sketch their app design and then generate code for various platforms such as SwiftUI, Flutter, Kotlin Jetpack Compose, and Next.js. [Glowby] Explore more of our coverage of: AI in Software Development, GenAI Assistants, Code Generation Tools. Share this story by email