Amazon's AI Pledge

In this week's AI roundup, we witness a convergence of cutting-edge technologies and responsible development practices that are shaping the future of artificial intelligence. From Amazon's commitment to ethical AI to Meta's ambitious plans for Llama 3, the industry continues to push boundaries while addressing crucial concerns about safety and transparency. Anthropic's enhancements to Claude 3.5 Sonnet increase the token size, Microsoft's novel approach to spreadsheet analysis, and advancements in GPU utilization for language models highlight the ongoing quest for improved performance and efficiency. Additionally, Google's Project Oscar and Mistral's specialized AI models demonstrate the growing application of AI in software development and specialized domains. These developments collectively paint a picture of an AI landscape that is not only advancing rapidly but also evolving to meet complex challenges and diverse needs across industries.

Let's dive into these groundbreaking AI developments shaping our technological landscape.

Stuff you should know

Amazon's Commitment to Responsible AI Development

Amazon has joined a collaborative effort to promote responsible AI development, working with the White House and other stakeholders. They've integrated safety and ethical considerations into their AI processes, launched numerous responsible AI features, published extensive research, and provided comprehensive employee training. Amazon advocates for transparency in AI development and deployment, emphasizing the importance of collaboration between public and private sectors. They support establishing AI guidelines that align with democratic values, promote economic growth, and ensure global compatibility while fostering safe innovation.

Meta Planning to Launch Llama 3 400B Model

Meta is set to release its most powerful AI language model yet, Llama 3 400B, by the end of July 2024. This highly anticipated model boasts over 400 billion parameters and is expected to achieve near-parity with OpenAI's GPT-4 on the MMLU benchmark, despite using less than half the parameters. If open-sourced under the same license as previous Llama models, it would democratize access to state-of-the-art language capabilities, allowing researchers and developers to leverage this powerful tool for their projects without relying on expensive proprietary APIs. However, there is some uncertainty surrounding the release, as a recent tweet suggests Meta may not open-source the weights for the 400B model, contradicting previous statements and raising questions about the model's future accessibility.

Anthropic Doubles Claude 3.5 Sonnet's Token Limit

Anthropic has doubled the maximum output token limit for Claude 3.5 Sonnet from 4,096 to 8,192 tokens. This significant increase allows for much longer responses and the ability to handle more complex and extensive tasks and conversations. The 8,192 token limit is currently only available as a beta feature in the Anthropic API and console. However, given Anthropic's track record of rapid development and deployment, it's likely this expanded limit will soon be rolled out to the claude.ai platform as well. The higher token limit enables Claude 3.5 Sonnet to generate more detailed, comprehensive, and in-depth outputs. This is particularly beneficial for use cases requiring lengthy text generation, such as long-form writing, detailed analysis, and extended dialogues.

SPREADSHEETLLM: Excel's AI Translator

SPREADSHEETLLM is a novel framework developed by Microsoft researchers that aims to effectively leverage large language models (LLMs) for understanding and analyzing spreadsheet data. It includes an efficient encoding method called SHEETCOMPRESSOR, which consists of three modules: structural-anchor-based extraction, inverted-index translation, and data-format-aware aggregation. These modules significantly reduce token usage by up to 96% while preserving critical information, enabling LLMs to comprehend both structured and unstructured data within spreadsheets, including multiple tables and varied data formats. By using spreadsheets as a "source of truth", SPREADSHEETLLM may reduce AI hallucinations and improve the reliability of AI outputs, especially in areas like financial analysis and data science.

GPU Turbocharge: FlashAttention 3 Supercharges LLMs

FlashAttention 3, a novel AI technology, has been released to unleash the power of NVIDIA H100 GPUs for large language models (LLMs). This breakthrough enables LLMs to achieve unprecedented performance and efficiency by leveraging the advanced capabilities of H100 GPUs. FlashAttention 3 introduces several key innovations, including optimized attention computation, memory-efficient training, and dynamic tensor rematerialization. These advancements allow LLMs to process information more effectively, leading to faster training times and improved accuracy. By harnessing the parallel processing power of H100 GPUs, FlashAttention 3 can significantly accelerate the development and deployment of state-of-the-art LLMs, driving progress in natural language processing and generation tasks. This technology represents a significant step forward in the field of AI, paving the way for more advanced and capable language models that can tackle complex problems and assist humans in a wide range of applications.

Project Oscar Google's AI Platform for Software Development

Google has introduced Project Oscar, an open-source platform that allows software development teams to employ and build AI agents to manage various aspects of their projects, such as monitoring bugs and engaging with contributors. The AI agents in Project Oscar can interact using natural language, enabling developers to provide instructions without needing to rewrite code. Project Oscar was first deployed within Google's Go programming language project, and the AI agents can enhance issue reports by analyzing data and engaging directly with issue reporters. While currently focused on open-source projects, Google plans to extend Project Oscar to manage closed-source projects as well.

Mistral Unleashes AI Twins: Code Whiz and Math Genius

Mistral, a French AI startup, has released two new large language models - Codestral Mamba 7B and Mathstral 7B. Codestral Mamba is a code-generating model designed for programmers and developers, leveraging the efficient Mamba architecture to provide faster responses and handle longer input texts. Comparative evaluations showed Codestral Mamba outperforming competing open-source models. Mathstral 7B is a specialized AI model for mathematical reasoning and scientific exploration, which Mistral claims outperformed existing models in benchmark tests. Both models are available on Mistral's platforms and can be fine-tuned by users. Mistral's commitment to open-source principles has positioned it as a competitor to industry giants, and the company recently secured $640 million in Series B funding.

Hand Picked Video

In that video we can see Anthropic's recent update to Claude AI, users can now publish and remix artifacts created during their interactions with the AI assistant.

Top AI Products from this week 

• Phaie AI by Creatr - Phaie AI is an open-source tool you can use to generate and fix design systems. Using our Figma plugin, you can also work in existing files to create styles using AI to detect and change different colors, typographies, strokes, border radii, and much more.

• GPT Power Manager - Supercharge your ChatGPT experience with advanced organization and management tools.

• AI Web Designer - AI Web Designer allows users and private companies to redesign their websites in seconds using our AI model. They can always edit and export the design.

• AI image extender - Click and drag to expand beyond the edges of your image, to enlarge the background, adjust the aspect ratio, and more.

• PngMaker.io: Convert Text to PNG Quickly - PngMaker.io is a free online tool that converts text to PNG images with transparent backgrounds in seconds. Perfect for web design, marketing, and digital content.

• Sugar Free: Food Scanner - Detect hidden sugars with our label scanner • Get personalized guidance from an AI nutritionist • Track your progress and beat sugar addiction • Ultimate lifestyle tool for a healthier you!

• Project Atlas Desktop - Project Atlas generates complex business automation agents step-by-step, then it outputs a no-code drag-drop graph for you to edit.

• Jobright AI - Discover your dream job with Jobright, your AI job search copilot. Find relevant opportunities tailored to your goals.

• Blaze - Blaze uses AI and automation to scan social posts, engagements and connections and help companies target the right potential customers with more compelling outreach as soon as they show interest on channels like X and LinkedIn.

This week in AI

• AI Giants Reshuffle Board Seats Amid Scrutiny - Microsoft, Apple, and OpenAI have made changes to their board observer seats amid growing concerns over AI risks and ethics. Microsoft and Apple have stepped down from their observer roles on OpenAI's board, while OpenAI has lost its regulator seat on the OpenAI Foundation's board. These moves signal a shift in the companies' priorities and strategies in response to recent controversies and increased regulatory scrutiny of AI development.

• Kling Introduces Viral Short Video Feature in China - Landing AI's text-to-video model Kling has launched a new short video generation feature in China, which has gone viral. The feature allows users to generate short videos from text prompts, showcasing the rapid advancements in AI-powered content creation.

• Anthropic Unleashes Claude on Android - Anthropic has launched its first official Android app for its conversational AI assistant Claude. The app offers natural language conversations, task assistance, and creative collaboration, with features like offline mode and image analysis.

• YouTube Music Tests AI-Generated Radio and Adds Song Recognition - YouTube Music is experimenting with an AI-generated radio feature and introducing a song recognition tool. The AI radio aims to provide personalized music recommendations by analyzing users' listening habits and preferences.