Together with

Today:

Stay up-to-date with AI.

AI won’t replace you, but a person using AI might. That’s why 500,000+ professionals read The Rundown– the free newsletter that keeps you updated on the latest AI news, tools, and tutorials in 5 minutes a day.

Sign up today.

Building $1,000,000+ AI Apps | a16z Investment Thesis "The Abundance Agenda" | Build, Invest or Use.

a16z, a big-shot venture capital firm led by internet mogul Mark Andreessen, is all in on AI, betting that today's high-end tech will be tomorrow's everyday stuff. The firm's "Abundance Agenda" suggests a future where AI makes high-end goods and services more accessible to everyone. They're essentially telling tech innovators: if you build AI apps, they're ready to fund them. 

With examples like MultiOn showcasing AI's potential to handle complex tasks, a16z's roadmap is clear: AI is not just the future; it's the present, and it's reshaping the tech landscape in big, bold ways.

WATCH THE VIDEO ON YOUTUBE

OpenAI forms a new team to study child safety

OpenAI is forming a team to tackle child safety concerns with its AI tools, responding to worries from activists and parents. The new team aims to prevent misuse by kids, focusing on policies, investigations, and reviews. OpenAI is hiring a specialist to enforce child safety policies regarding AI-generated content. 

This move aligns with industry norms, where tech companies allocate resources to comply with regulations like the U.S. Children’s Online Privacy Protection Rule. The initiative reflects OpenAI's proactive stance amid growing concerns about minors using AI tools for both academic and personal purposes, emphasizing the need for clear guidelines and safeguards.

TECHCRUNCH

Microsoft brings new AI image functionality to Copilot, adds new model Deucalion

Microsoft unveiled a revamped Copilot AI with improved image features and a new model called Deucalion. The redesign offers a cleaner interface with more visuals. The AI can now generate images and edit them directly within Copilot, targeting users in various English-speaking countries. 

Microsoft aims to boost Copilot's popularity through a Super Bowl ad, showcasing its ability to create content beyond just search. Despite concerns about AI misuse, Microsoft moves forward, introducing Deucalion, a fine-tuned model for Copilot's Balanced mode. Its origins are unclear, but it likely builds upon GPT-4.

VENTUREBEAT

Apple releases ‘MGIE’, a revolutionary AI model for instruction-based image editing

Apple has launched "MGIE," an innovative AI model for editing images based on natural language instructions. Developed in collaboration with UC Santa Barbara researchers, MGIE uses large language models (LLMs) to interpret user commands and perform detailed edits like Photoshop-style modifications, global photo enhancements, and local edits. It integrates LLMs into the editing process by deriving clear instructions from user input and generating a visual representation of the desired edit. 

MGIE is available as an open-source project on GitHub, providing users with a flexible and easy-to-use tool for various editing tasks. This breakthrough highlights Apple's AI advancements and signifies a promising future for assistive AI in creative endeavors.

VENTUREBEAT

Google showcases AI accessibility in a new ad.

Google highlights AI accessibility in a fresh ad. The Pixel 7's Guided Frame, using AI, narrates selfie camera content. The Super Bowl 2024 commercial, featuring director Adam Morse, who is visually impaired, demonstrates its value for those with low vision.

THE VERGE

Microsoft CEO Satya Nadella Wants To Train 2 Million People In India With AI Skills

Microsoft CEO Satya Nadella aims to train 2 million people in India in artificial intelligence by 2025. The initiative targets Tier 2 and Tier 3 cities as well as rural areas, collaborating with governments and organizations. Nadella, emphasizing India's potential in AI, believes investing in the technology will drive economic growth, likening it to past transformative investments. 

Amid concerns about AI's impact on jobs and inequality, Microsoft's move aligns with its broader AI integration strategy. The announcement follows Microsoft's recent valuation milestone and previous commitments to AI development in India.

FORBES

🧠RESEARCH

MusicRL: Aligning Music Generation to Human Preferences

Google DeepMind's MusicRL is the first music creation system tweaked using people's feedback to get better. It learns from what listeners like, making tunes that hit closer to home, whether you're after some pump-up jams or chill vibes. By using real feedback, it's shaping up to crank out hits that feel just right, showing that what we dig in music is super personal and there's still a ton to learn about making beats that please our ears.

MetaVoice-1B

MetaVoice-1B is an advanced AI for creating human-like speech, trained on 100K hours of speech. It excels in emotional speech rhythm and tone in English, supports cross-lingual voice cloning with minimal training data, and offers zero-shot cloning for American & British voices. It's released under the Apache 2.0 license, allowing unrestricted use. The system is designed for both emotion-rich speech and long-form content synthesis, with a demo available for testing its capabilities.

IMUSIC: IMU-based Facial Expression Capture

IMUSIC introduces a novel approach to capturing facial expressions using IMU (Inertial Measurement Units) signals, a departure from traditional visual-based methods. This technique, focused on privacy and overcoming limitations like occlusions, employs micro-IMUs tailored for facial movements, paired with an anatomy-driven placement scheme. It leverages a unique IMU-ARKit dataset for training, leading to accurate facial blendshape parameter prediction.

Diffusion World Model

 Diffusion World Model (DWM) is a breakthrough in predictive modeling for reinforcement learning. Unlike traditional models that predict future states one step at a time, DWM can forecast multiple future states and rewards in a single pass. This enables more accurate long-term predictions without recursive calculations. Applied to offline reinforcement learning, DWM enhances value estimation with its generative modeling approach, showing significant performance improvements over one-step models in experiments.

Self-Discover: Large Language Models Self-Compose Reasoning Structures

"Self-Discover" introduces a new way for Large Language Models (LLMs) to solve tough reasoning problems by figuring out their own reasoning structures, combining basic reasoning modules like critical and step-by-step thinking. This method significantly boosts the performance of models like GPT-4 and PaLM 2 on hard reasoning tasks, outperforming previous techniques while using far less computing power. It also reveals that these self-made reasoning patterns are similar to human reasoning and work across different LLMs. 

EVA-CLIP-18B: Scaling CLIP to 18 Billion Parameters

EVA-CLIP-18B scales up the CLIP model to an impressive 18 billion parameters, setting a new benchmark for open-source vision and multimodal models. Achieving an 80.7% zero-shot top-1 accuracy across 27 image classification benchmarks, it outperforms previous models with fewer training samples. This model, trained on a dataset significantly smaller than those used by other state-of-the-art models, showcases the potential of scaling up visual models for improved performance. With its public release, EVA-CLIP-18B aims to advance research in vision and multimodal foundation models. 

🛠️TOP TOOLS

Librarian - facilitates data management and organization, making it user-friendly for various data-related tasks.

Omnipilot - provides AI-driven solutions and services. It focuses on harnessing the power of artificial intelligence for diverse applications and industries

Rask AI - Translate audio and video from any language into over 130+ languages.

Moonbeam - AI writing assistant designed for creating various types of content. It offers templates, Smart Chat, content clusters, and a powerful text editor.

cre8tiveAI - online platform revolutionizing creative retouching work, offering a suite of AI tools for photo and video editing. 

📲SOCIAL MEDIA

🗞️MORE NEWS

Microsoft’s next big AI push is here after a year of Bing

Microsoft shifts focus from Bing to Copilot, marking a year since AI-powered Bing launch. Despite Google's dominance, Microsoft sees growth in Bing. Copilot, now central to Microsoft's AI strategy, emphasizes productivity over search. Super Bowl ad promotes Copilot as AI companion for creativity. Microsoft aims for unified Copilot experience across products. THE VERGE

World’s biggest AI tech companies push UK over safety tests

The UK government is being urged by major AI companies, including OpenAI, Google DeepMind, Microsoft, and Meta, to speed up safety tests for AI systems. These companies signed voluntary commitments in November to allow review of their AI models by the AI Safety Institute (AISI) in Britain. The companies are seeking clarity on the tests being conducted, their duration, and the feedback process if any risks are found. The government has outlined the need for future binding requirements to ensure AI developers are accountable for keeping systems safe. FINANCIAL TIMES

Stability AI launches SVD 1.1, a diffusion model for more consistent AI videos

Stability AI releases SVD 1.1, an improved version of its AI model for generating consistent short videos. Available for public use via Hugging Face, with subscription options for commercial deployment. The upgrade promises better motion and reliability compared to previous versions, addressing past limitations. Competitors like Runway and Pika offer similar services but lack API integration. VENTUREBEAT

US agency to hire 50 AI experts to crack down on drugs, child abuse

The U.S. Department of Homeland Security (DHS) plans to hire 50 AI experts to combat child abuse, counter fentanyl production, and assess natural disaster damage. This move reflects the Biden administration's push to leverage AI benefits while addressing potential threats. AI is already aiding DHS in detecting illegal drugs and assisting disaster response efforts. REUTERS

Meet ‘Smaug-72B’: The new king of open-source AI

The new open-source language model, Smaug-72B, surpasses GPT-3.5 and Mistral Medium, marking a significant advancement in AI. Developed by Abacus AI, it excels in natural language understanding tasks and sets a benchmark for open-source AI. Its release signals a shift towards democratizing AI innovation and challenging Big Tech's dominance. VENTUREBEAT

Health Systems Are Improving Patient Outcomes With AI-Assisted Technology

AI technology has the potential to reduce clinician burden and improve patient experience in healthcare. Hospitals are already using AI to improve patient outcomes, such as in the case of sepsis diagnosis and treatment. Using AI to catch sepsis sooner led to a 48% increase in CMS bundle compliance, resulting in better reimbursement for hospitals. AI is not meant to replace clinicians, but rather to empower them and make them more efficient and effective. FORBES

AI Predicts Antidepressant Success in a Week

Researchers developed an AI algorithm to predict antidepressant efficacy for major depression disorder patients within a week, potentially reducing unnecessary prescriptions of sertraline. By analyzing brain scans and symptoms, it identifies responders faster, offering improved care and societal savings. The approach personalizes treatment, marking a significant medical advancement. NEUROSCIENCE NEWS

AI-backed algo crafts 3D printed foam, boosts helmet protection by 25%

Researchers from the University of Colorado of Boulder and Sandia National Laboratories have developed a novel helmet padding design using AI, enhancing impact absorption by up to 25%. Their 3D-printable foam, with controlled buckling patterns, outperforms conventional materials, potentially revolutionizing safety gear for sports and transportation. INTERESTING ENGINEERING

AI is supercharging collaboration between developers and business users

AI is revolutionizing collaboration between developers and business users, with tools like generative AI, GitHub Copilot, and Azure Machine Learning Studio aiding in code generation and testing. As AI adoption grows, it fosters closer teamwork, enhances DevOps processes, and boosts innovation, ultimately reshaping software development workflows. ZDNET

The best AI image generators to try right now

AI-powered image generators are revolutionizing creative expression. In a world where visuals play a pivotal role in communication and branding, these tools offer instant gratification, transforming text prompts into stunning visuals. From industry giants like Microsoft and Google to innovative newcomers, we uncover the top players reshaping the way we visualize ideas.

Image Creator from Microsoft Designer (formerly Bing Image Creator)

• Powered by DALL-E 3

• Quick and accessible via Copilot, browser, and mobile

• Free of charge

DALL-E 2 by OpenAI

• Fast and intuitive

• Offers accurate depictions

• Free with credits or subscription

ImageFX by Google

• Easy-to-use with high-quality results

• Features expressive chips for prompt refinement

• Free of charge

DreamStudio by Stability AI

• Allows for deep customization

• Open-source with various customization options

• Offers 25 free credits upon account creation

Dream by WOMBO

• Mobile-friendly with multiple templates

• Realistic renditions with a range of design styles

• Offers a free limited access version

Craiyon

• Simple and straightforward interface

• Unlimited prompts for experimentation

• Free of charge

Midjourney

• Produces highest quality photos

• Features a Discord community for sharing

• Starts at $10/month

Generative AI by Getty Images

• Generates commercially safe images

• Utilizes content solely from Getty Images' library

• Pricing determined through contact with the sales team

ZDNET