In partnership with

Good morning. It’s Friday, April 5th.

Did you know: On this day in 1975, Microsoft was founded in Albuquerque, New Mexico?

In today’s email:

• Claude 3 outperforms GPT-4 in text summarization

• Microsoft and Quantinuum reach key quantum computing milestone

• Microsoft 365's Copilot upgraded with GPT-4 and better image generation

• YouTube CEO's caution to OpenAI on Sora data may backfire

• 10 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

^{In partnership with MaxAI Me}

MaxAI.me - Outsmart Most People with 1-Click AI

Discover MaxAI.me, one of the top 50 GenAI apps of 2024!

Best features:

• Chat with the latest AI like GPT-4, Claude 3, and Gemini 1.5, all in one place.

• Perfect your writing anywhere with 1-click AI without copy-pasting.

• Save 90% of your reading & watching time with AI summaries.

• Reply 10x faster with AI on email, social media, and messaging web apps.

• Rapidly turn your visions into stunning images with AI art generators.

Join 1M+ users on MaxAI.me to get more done in less time!

Anthropic's Claude 3 Beats OpenAI's GPT-4 at Text Summarization

A new dataset, Faithfulness Annotations for Book-Length Summarization (FABLES), offers insights into AI summarization of entire books.

Researchers evaluated summaries generated by different models, finding Anthropic's Claude 3 Opus to be more faithful than OpenAI's GPT-4, with 90% of annotations deemed accurate, though verifying factual claims automatically remains a challenge. The study reveals that longer context windows, while reducing chronological errors, lead models to prioritize content from book endings.

Claude 3 Opus demonstrates strength in including key information but struggles with perfect summarization of lengthy texts. The FABLES dataset is available on GitHub for further exploration of AI-powered summarization. Read more.

Microsoft and Quantinuum Achieve Major Milestone in Reliable Quantum Computing

Quantinuum, in partnership with Microsoft (+4.07 (0.97%) today), has achieved a significant advancement in quantum computing by demonstrating the most reliable logical qubits to date.

Enabled by the synergy between Quantinuum's H2 quantum processor and Microsoft's qubit-virtualization system, this breakthrough highlights logical qubits with error rates 800 times lower than physical error rates.

This development, previously projected to be years away, has the potential to accelerate solutions to real-world problems in fields such as materials science and drug discovery. Leveraging the outstanding fidelity, scalability, and flexibility of Quantinuum's H2 processor, powered by Honeywell, and Microsoft's innovative error correction capabilities, the collaboration sets a new standard in quantum computing reliability. Read more.

Microsoft 365’s Copilot Gets GPT-4 Turbo Upgrade and Improved Image Generation

Microsoft upgrades Copilot for business users by integrating OpenAI's GPT-4 Turbo model, offering unlimited daily usage and advanced AI capabilities across platforms (mobile, web, Windows, Edge).

Included in the $30/user/month Copilot for Microsoft 365 subscription, GPT-4 Turbo expands functionality with features like in-depth document analysis (up to 300 pages). Additionally, Microsoft enhances Designer with OpenAI's DALL-E 3, enabling the generation of up to 100 images per day, further boosting productivity for subscribers. Read more.

YouTube CEO's Warning to OpenAI: Sora Training Data Could Backfire “Spectacularly”

YouTube CEO Neal Mohan has thrown a wrench into AI data sourcing practices.

He suggests OpenAI's text-to-video model, Sora, might be trained on YouTube videos in violation of the platform's terms. This ignites a debate on creator rights and legal implications.

Creators, Mohan argues, expect their work to be respected, and YouTube's terms forbid using content for purposes beyond viewing.

This directly challenges the practice of "scraping" web data, including copyrighted material, to train AI models like Sora. The situation becomes particularly sticky for Google, facing lawsuits over allegedly using unauthorized data for its own AI. Google defends its actions as "fair use," claiming the data is transformed during training. Read more.

🖇️ Etcetera

> Deciphering genomic language: New AI system unlocks biology's source code

> FDA authorizes Prenosis software as first AI tool that can diagnose sepsis

> Assembly AI claims its Universal-1 model generates 30% fewer transcription errors than OpenAI's Whisper Large-v3 model

> Cohere releases Command R+ AI model designed for enterprise-scale use

> Business Schools Add AI to Curriculum

> Codium announces Codiumate, a new AI agent that seeks to be Devin for enterprise software development

> Hackers force AI chatbots to break their own rule

> Google plans to charge for AI-powered search engine, FT reports

> Opera becomes the first major browser with built-in access to local AI models

> Elon Musk says he's raising Tesla engineer salaries because OpenAI has been aggressively poaching them with massive paydays

> Vimeo Hires Google Cloud AI Exec Philip Moyer as CEO

> Former Snap AI chief launches Higgsfield to take on OpenAI’s Sora video generator

> Google hires former OpenAI executive to boost AI efforts

AITalk enhances language learning on iOS with AI-driven tutoring. Engage in real-time conversations, access diverse languages, and utilize advanced learning tools, including personalized IELTS prep.

Fracttal One is an AI-driven maintenance management tool in LatAm. It has a modern interface and adaptive functionalities. Integrates AI assistants and predictive maintenance.

​​Litespace is an all-in-one employee experience platform enhancing engagement, culture, and efficiency. Offers AI-powered tools for team management, streamlining communication, and dynamic analysis.

Living Images uses genAI for effortless image optimization. Input your site, select images, and integrate our script or image URL. AI iteratively learns and generates improved images over time.

IntelSwift is a customer service AI with human decision-making capabilities. Enhance CX throughout the customer journey with AI chatbots, AI Copilot, Live Chat, and Smart Analytics.

Metaforms AI combines OpenAI and Typeform for feedback and research featuring personalized questions and deeper insights.

Noctie.ai offers personalized chess practice with human-like play and instant feedback, catering to players of all levels. Learn openings and receive custom exercises.

SAAS GPT, developed by Quest Labs AI, swiftly creates personalized React SaaS dashboard starters, offering customization options and integration with analytics.

Release AI automates API change tracking and generates accurate standardized release notes. Free tier available, with a Pro plan giveaway.

BannerGate.ai simplifies animated HTML5 display banner creation for advertising campaigns. Its AI-driven platform offers customization without coding.

📄 Freditor: High-Fidelity and Transferable NeRF Editing by Frequency Decomposition

📄 Language Models as Compilers: Simulating Pseudocode Execution Improves Algorithmic Reasoning in Language Models ​​

📄 Mixture-of-Depths: Dynamically allocating compute in transformer-based language models

📄 ChatGLM-Math: Improving Math Problem-Solving in Large Language Models with a Self-Critique Pipeline

📄 InstantStyle: Free Lunch towards Style-Preserving in Text-to-Image Generation

📄 RALL-E: Robust Codec Language Modeling with Chain-of-Thought Prompting for Text-to-Speech Synthesis

📄 MiniGPT4-Video: Advancing Multimodal LLMs for Video Understanding with Interleaved Visual-Textual Tokens  

📄 PointInfinity: Resolution-Invariant Point Diffusion Models

📄 ReFT: Representation Finetuning for Language Models  

📄 Training LLMs over Neurally Compressed Text

AI Creates Comics

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, apply here.