In partnership with

Good morning. It’s Wednesday, April 24th.

Did you know: 40 years ago today, the iconic Apple IIc was released?

In today’s email:

• Microsoft’s Phi-3

• Google’s Prompt Guide

• Adobe Firefly Now Free

• Ray Ban’s Multimodal AI Glasses

• 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}

Learn AI in 5 Minutes a Day

AI Tool Report is one of the fastest-growing and most respected newsletters in the world, with over 550,000 readers from companies like OpenAI, Nvidia, Meta, Microsoft, and more.

Our research team spends hundreds of hours a week summarizing the latest news, and finding you the best opportunities to save time and earn more using AI.

Sign up with 1-Click

Microsoft Launches Phi-3, Its Smallest AI Model Yet

Microsoft joins the lightweight LLM race with the Phi-3 Mini, a 3.8 billion parameter model designed for on-device performance. Trained on a curated dataset (think quality over quantity compared to GPT-4), Phi-3 targets resource-constrained scenarios and low-latency applications. Its training dataset, an expanded version of Phi-2's, includes heavily filtered web data and synthetic data, ensuring robustness, safety, and chat format optimization.

This is just the first glimpse of the Phi-3 family, with Phi-3 Small (7B) and Phi-3 Medium (14B) on the horizon.

Phi-3 Mini punches above its weight, rivaling models ten times its size in performance. The Phi-3-Mini achieves impressive results (69% on MMLU comprehension and 8.38 on MT benchmarks) while running locally on smartphones with a memory footprint quantized to 4 bits and occupying around 1.8GB of memory. It dives into the sub-4B parameter LLM market enabling on-device AI and faster dev cycles.

These smaller, efficiently trained models pave the way for future personalized AI experiences directly on devices. Notable examples include EleutherAI's open-sourced GPT-Neo with 2.7 billion parameters, Google's Gemma 2B, and Huawei's TinyBERT.

Read more.

Related Links: Performance Comparison on Selected Tasks

Paper: Phi-3 Technical Report

Google releases New Free Resources to Help You Optimize Your AI Prompts

Google released a blog series and an ebook, Prompting guide 101 to improve user experience with AI chatbots. The resources focus on effective prompt design for generative AI models like Google's Gemini.

A well-structured prompt considers four key elements: persona (information requester), task (desired action), context (additional details), and format (output structure).

The "Prompting guide 101" ebook offers general prompting advice and role-specific prompts for marketing, customer service, and other professions. These resources aim to enhance user ability to leverage AI chatbots across various platforms. Read more.

Adobe’s latest AI Image Generator Is Now Free

Adobe's latest AI model, Firefly 3, is now accessible via the Firefly web app, offering enhanced accuracy and photorealistic results. Forget app hopping - this update seamlessly integrates full AI image generation within Photoshop thanks to Firefly 3.

Users can now whip up hyper-realistic visuals directly using text prompts or pre-trained settings. This injects serious creative firepower, leveraging Photoshop's massive toolbox (70+ tools) for fine-tuning.

New features like Reference Image and Generate Background push the creative envelope further, making marketing asset creation a breeze. Ethical AI stays a focus, with training on licensed datasets and IP indemnification for generated content. Read more.

The Ray-Ban Meta Smart Glasses Have Multimodal AI Now

The Ray-Ban Meta smart glasses just got a serious upgrade, following a successful early access program. Unlike the ill-fated Humane AI Pin, Meta's AI assistant shows real promise in the wearables market.

Users can now initiate commands with a simple "Hey Meta, look and..." to perform tasks like identifying objects or translating languages on the fly. Despite occasional imperfections, the AI proves practical for everyday usage, particularly outdoors.

The AI complements other features like live streaming and POV camera capabilities. While there are still occasional quirks to iron out, the addition of multimodal AI brings significant value to this already impressive product. Read more.

🖇️ Etcetera: Stories You May Have Missed

> OpenAI shrugs off Llama 3's ascent with new enterprise AI features 

> Apple quietly beefs up its AI division with acquisition of French startup

> Smartwatch AI predicts atrial fibrillation 30 minutes before it arrives

> AI tool creates 'synthetic' images of cells for enhanced microscopy analysis

> Quantum Computers Can Run Powerful AI That Works like the Brain

> Scientists create 'toxic AI' that is rewarded for thinking up the worst possible questions we could imagine

> With a game show as his guide, researcher uses AI to predict deception

> AI Learns to Bluff Like a Gameshow Champ

> Microsoft rolls out authentication tools to combat AI-powered election disinformation

> Video: Inside the World's First Dogfight Between AI and Human Pilot

> Google DeepMind RecurrentGemma Beats Transformer Models

> A new framework to generate human motions from language prompts

> AI Starts to Sift Through String Theory’s Near-Endless Possibilities 

> Apple Reportedly Developing Its Own Custom Silicon for AI Servers

> AI search engine Perplexity.ai hits billion dollar valuation and releases new enterprise offering

Chatbot Arena offers side-by-side comparison of features, training, and pricing so you can find the best chatbot builder.

Vidnoz AI 2.8 offers a free Synthesia alternative with video translation. Features include 600+ AI avatars, 470+ voices, and 700+ templates.

SQLyze simplifies SQL query creation with AI assistance. It remembers tables, fields, and content examples, enhancing database management efficiency. Free platform.

UnitWise is an AI-powered platform, assists developers in generating unit tests within IDEs. It offers insights, auto-fixes, and customizable features. Supports multiple languages. Free options available.

HitPaw Video Editor is an AI-powered tool for both amateur and professional editors. It simplifies editing with features like auto-subtitles and AI-driven tools.

Insights by Ayraa introduces AI-driven enterprise search, expanding beyond text to delve into workplace data for nuanced insights. Explore productivity enhancement. Free with optional Pro features.

Intrvu Space streamlines hiring with AI-driven interviews, report generation, and approval automation. Conduct conversational interviews resembling online meetings.

Voxal.AI offers an AI chatbot solution going beyond text, enhancing engagement, branding, and conversions. Compatible with any website, including Shopify, Wordpress.

Jamie AI simplifies meeting documentation with AI, automatically generating summaries, transcripts, and action items from audio recordings. Supports 15+ languages and respects privacy.

ProDream is an AI-driven academic writing tool. It offers outlining, editing, citation, and plagiarism checks. Access 200M scholarly sources, multilingual support.

📄 SEED-X: Multimodal Models with Unified Multi-granularity Comprehension and Generation

📄 FlowMind: Automatic Workflow Generation with LLMs ​​

📄 The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions

📄 Learning H-Infinity Locomotion Control  

📄 Transformers Can Represent n-gram Language Models

📄 Pegasus-1 Technical Report

📄 Align Your Steps: Optimizing Sampling Schedules in Diffusion Models

📄 OpenELM: An Efficient Language Model Family with Open-source Training and Inference Framework 

📄 Multi-Head Mixture-of-Experts  

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.