Want to get the most out of ChatGPT?

Revolutionize your workday with the power of ChatGPT! Dive into HubSpot’s guide to discover how AI can elevate your productivity and creativity. Learn to automate tasks, enhance decision-making, and foster innovation, all through the capabilities of ChatGPT.

Download the offer for actionable insights and tips on harnessing AI for a more efficient, impactful professional life. 

Today:

• OpenAI Launches Cheaper GPT-4o

• Claude Android App Now Available

• Google Launches Vids AI App

• Mistral Unveils Codestral Mamba 7B

• Eureka Labs Launches AI School

• Spotify Introduces Spanish AI DJ

• New Creative Tools in Microsoft Designer

• $100M Anthropic AI Startup Fund

OpenAI unveils GPT-4o mini — a smaller, much cheaper multimodal AI model

OpenAI has launched GPT-4o mini, a smaller and cheaper version of its multimodal AI model, GPT-4o. Costing developers $0.15 per million input tokens and $0.60 per million output tokens, it is significantly less expensive than its predecessor. GPT-4o mini handles text, image, audio, and potentially video inputs and outputs. 

It aims to be a cost-efficient solution for enterprises and developers creating various applications. Although it will replace GPT-3.5 Turbo in ChatGPT, the older model will still be supported. OpenAI plans to integrate GPT-4o mini into Apple devices this fall.

VENTUREBEAT

Claude Android app

Anthropic has released the Claude Android app, offering access to their advanced AI model, Claude 3.5 Sonnet. The app is free for all plans, including Pro and Team, and mirrors the functionality of Claude on iOS and the web. Users can engage with Claude across multiple platforms, utilize vision capabilities for real-time image analysis, and benefit from multilingual processing for language translation. 

The app supports advanced reasoning tasks, aiding in activities such as contract analysis, market research, and speech composition. The Claude Android app is available for download on Google Play.

ANTHROPIC

Google tests out Gemini AI-created video presentations

Google has introduced Vids, a new AI-powered productivity app in Workspace Labs. Announced in April, Vids simplifies video creation by allowing users to integrate documents, slides, voiceovers, and recordings into a timeline. The AI, Gemini, helps by automatically inserting stock footage, generating scripts, and providing AI voiceovers. 

Unlike tools that create lifelike footage from prompts, Vids focuses on creating presentation videos. Available in the Workspace Labs preview, Workspace admins can enable user access to try it out. The app also offers templates and royalty-free content to enhance presentations.

THE VERGE

Mistral releases Codestral Mamba for faster, longer code generation

French AI startup Mistral launched two new models: Codestral Mamba 7B for code generation and Mathstral 7B for math-related tasks. Based on the efficient Mamba architecture, Codestral Mamba excels in code productivity, handling up to 256,000 tokens, and outperforms competitors in HumanEval tests. It’s free on Mistral’s la Plateforme API and GitHub under an Apache 2.0 license. 

Mathstral, designed for STEM applications, boasts a 32K context window and superior performance in math reasoning. Both models highlight Mistral’s commitment to open-source AI and specific-purpose development. Mistral recently raised $640 million, reaching a valuation close to $6 billion.

VENTUREBEAT

Introducing Eureka Labs

Eureka Labs is creating an AI-native school to enhance learning experiences. Their approach combines expert-designed courses with AI Teaching Assistants, providing personalized guidance for students. This AI support can scale education, making it accessible and efficient for a global audience. Their first course, LLM101n, will teach students to train their own AI. 

Available online and in cohorts, this undergraduate-level class exemplifies their vision of expanding education through AI. Eureka Labs aims to harness AI to boost human potential, making learning more accessible and extensive. Follow them on social media or join their newsletter for updates.

EUREKA LABS

Spotify launches a new voice and language for its AI DJ

Spotify has introduced a Spanish-language version of its AI DJ feature, expanding its availability to select markets in Latin America and Spain. The new voice, provided by Spotify’s senior music editor Olivia “Livi” Quiroz Roa, allows users to switch between her and Xavier “X” Jernigan, who voices the English-speaking DJ variant. 

This feature offers radio-like commentary with additional context about songs, paired with an endless playlist of personalized music recommendations that adapt to user activity. Available to Premium users, this rollout comes after recent price increases, which Spotify says will enable continued investment and innovation in their product features.

THE VERGE

New ways to get creative with Microsoft Designer, powered by AI

Microsoft Designer, an AI-powered design tool, is now available for personal Microsoft accounts, offering new features to enhance creativity. Users can describe their ideas, and Designer will generate images and designs. Integrated with Microsoft products like Word and PowerPoint, Designer helps maintain creative flow. It supports over 80 languages and is available as a free mobile app. 

New features include prompt templates, custom stickers, and personalized greeting cards. Users can also transform photos with AI-powered tools. Microsoft emphasizes responsible AI use, ensuring ethical and safe user experiences. Copilot Pro subscription offers additional boosts and enhanced capabilities.

MICROSOFT

Anthropic teams up with venture capital firm to kickstart $100M AI startup fund

Anthropic, in partnership with Menlo Ventures, has launched the Anthology Fund, a $100 million investment fund aimed at supporting AI startups. Each startup can receive at least $100,000, along with $25,000 in Anthropic credits and additional credits from Menlo Ventures' firms. The fund offers various benefits, including access to Anthropic’s AI models, expert guidance, and networking opportunities. 

Although use of Anthropic's AI model Claude is not required, it's anticipated. Anthropic's initiative parallels OpenAI's Startup Fund, reflecting both companies' trajectories. The fund aims to foster AI advancements in sectors like healthcare, legal services, and education.

THE REGISTER

🧠RESEARCH

Spectra: A Comprehensive Study of Ternary, Quantized, and FP16 Language Models

The paper introduces the Spectra LLM suite, comprising 54 models from 99M to 3.9B parameters, trained on 300B tokens. These include FloatLMs, QuantLMs, and improved ternary models (TriLMs). TriLMs match half-precision models in performance while being smaller. However, they show similar toxicity and stereotyping issues as larger models.

AgentPoison: Red-teaming LLM Agents via Poisoning Memory or Knowledge Bases

The paper introduces AgentPoison, a method to compromise LLM agents by poisoning their memory or knowledge bases. This backdoor attack requires no extra model training and is highly effective, achieving over 80% success in attacking various agents with minimal impact on normal performance. The approach raises concerns about the safety and trustworthiness of LLM agents.

Qwen2-Audio Technical Report

The Qwen2-Audio report details advancements in an audio-language model that processes audio inputs and provides analysis or text responses. It simplifies pre-training with natural language prompts and enhances instruction-following abilities. The model supports voice chat and audio analysis without needing system prompts for mode switching, outperforming previous state-of-the-art models in audio-centric tasks.

GRUtopia: Dream General Robots in a City at Scale

The paper introduces GRUtopia, a simulated 3D society for training robots. It features GRScenes, a dataset of 100k interactive scenes for city-scale environments, and GRResidents, an NPC system for social interaction and task management. The GRBench benchmark tests robots on tasks like Object Loco-Navigation and Loco-Manipulation, advancing Embodied AI research.

SpreadsheetLLM: Encoding Spreadsheets for Large Language Models

The paper presents SpreadsheetLLM, a novel method to enhance large language models' ability to understand and reason with spreadsheets. It introduces SheetCompressor, a framework that efficiently compresses spreadsheet data, improving performance by 25.6% in detection tasks and achieving a 78.9% F1 score in QA tasks, outperforming previous models.

🛠️TOP TOOLS

Microsoft Designer - Create unique designs based on your own photo powered by AI

Claude by Anthropic - Get help on a variety of tasks whenever inspiration strikes—on everything from writing to analysis to math—from Claude, your trusted AI assistant.

Synthesia - Create studio-quality videos with AI avatars and voiceovers in 130+ languages.

Buildbox - Create, design and build games with AI.

Haiper - Create video with text

🗞️MORE NEWS

The ‘godmother of AI’ has a new startup already worth $1 billion

Renowned computer scientist Fei-Fei Li, known as the "godmother of AI," has launched World Labs, a startup valued at over $1 billion in just four months. Backed by Andreessen Horowitz and Radical Ventures, World Labs aims to develop AI capable of advanced reasoning through human-like processing of visual data. Li, famous for her work in computer vision and the creation of ImageNet, aims to create AI that understands the three-dimensional physical world. THE VERGE

Apple, Nvidia, Anthropic Used Thousands of Swiped YouTube Videos to Train AI

AI companies, including Anthropic, Nvidia, and Apple, used subtitles from 173,536 YouTube videos to train AI models without creators' consent, violating YouTube's rules. This dataset, known as YouTube Subtitles, includes content from educational channels like Khan Academy, as well as popular creators like MrBeast and PewDiePie. Creators, such as David Pakman and companies like Salesforce, voiced concerns about the unauthorized use and potential exploitation of their work. PROOF

Meta Halts AI Use in Brazil Following Data Protection Authority's Ban

Meta has paused the use of generative AI in Brazil following a preliminary ban by the country's data protection authority. The ban stems from concerns over a new privacy policy that allows Meta to access users' personal data to train its AI systems. The Brazilian authority cited potential serious and irreparable harm to individuals' rights, imposing a daily fine for non-compliance. THE HACKER NEWS

Trump allies draft AI order to launch ‘Manhattan Projects’ for defense

Former President Trump's allies are drafting an AI executive order to develop military technology and reduce regulations, favoring Silicon Valley investors. This plan contrasts with Biden's administration, which mandates safety testing for AI systems. The proposed framework includes "industry-led" agencies to evaluate AI models and secure systems. The America First Policy Institute and the conservative Heritage Foundation are involved in drafting these policies. WASHINGTON POST

Ukraine rushes to create AI-enabled war drones

Ukraine is developing AI-enabled drones to gain a technological edge in warfare against Russia. Startups like Swarmer are creating AI systems to manage drone swarms, which can operate more effectively than human pilots. These drones use AI for target identification, navigation, and executing strikes with minimal human intervention.  REUTERS

AI method radically speeds predictions of materials’ thermal properties

MIT researchers have developed a machine-learning method that predicts materials' thermal properties up to 1,000 times faster than existing AI techniques. This innovation could lead to more efficient energy conversion systems and faster electronics by better managing heat, thanks to a new neural network framework that simplifies complex calculations. MIT