ML RUNDOWN: Check Out This Week’s AI Newsletter!

Welcome to Your AI News Update!

So much to explore in this week's discussions! From OpenAI’s new API features to Nvidia’s impressive AI model, many innovations are improving creativity and improving technology.

We’ll also touch on recent research exploring AI’s role in healthcare and introduce some new tools designed for creators.

Whether you’re a developer, a business professional, or just curious about AI, this newsletter will keep you informed about the exciting developments in the field.

Let’s get started!

| Latest AI News

OpenAI Announced 4 New APIs, and They’re Available Now?

OpenAI announced four major updates to its API services at DevDay, aimed at improving AI product development.?

These updates include:

1. Model Distillation – A tool for fine-tuning smaller models using larger ones, improving efficiency and performance.
2. Prompt Caching – Developers can save common prompts, reducing API costs by up to 50%.
3. Vision Fine-Tuning – GPT-4 can now be fine-tuned with images for enhanced visual applications like object detection and medical image analysis.
4. Realtime API – Enables faster, more responsive speech-to-speech applications without loss of emotion or accents, supporting functions like making appointments.

These updates will streamline AI development, lower costs, and improve performance across various applications.

Canvas by ChatGPT: What’s New?

OpenAI has launched Canvas, a new feature for ChatGPT that makes writing and coding projects easier. Instead of just chatting, users can work alongside ChatGPT in a separate window, where they can highlight sections, get suggestions, and make edits directly.

Key features of Canvas include:

• For writing: Users can adjust document length, change reading levels, polish grammar, and add emojis.
• For coding: Users can get help debugging, adding comments, and even converting code to different programming languages.

Currently, Canvas is available to ChatGPT Plus and Team users, with plans to expand access soon.

Read More!

Nvidia's New AI Model Rivals GPT-4 in Vision and Language Tasks

Nvidia has introduced NVLM 1.0, a new open-source AI model that can compete with big names like GPT-4. The top model, NVLM-D-72B, has 72 billion parameters and performs well with both text and images, even getting better at text-based tasks after training.

What makes this different is Nvidia's decision to make the model and its code available to everyone. This move could speed up AI development by giving smaller teams and researchers access to powerful tools usually kept private by big companies.

While this is a big step forward, it also raises concerns about how AI will be used and how businesses will adapt as advanced technology becomes easier to access.

For more in-depth details, read the full article.

Black Forest Labs, the startup behind Grok’s image generator, releases an API

Black Forest Labs, a startup supported by Andreessen Horowitz, launching a new API in beta for its image generation models called Flux. The API allows developers to integrate Flux models into their apps and services, with options like content moderation and image resolution limits.

Black Forest Labs also released a new, faster model called Flux1.1 Pro, which generates images six times faster than the previous version and can scale up to 2k resolution. This model is available on their platform and through partners like Together AI, Replicate, and Freepik.

Pricing for the service starts at 2.5 credits per image, with Flux1.1 Pro costing 4 credits per image. Black Forest Labs, co-founded by engineers who worked on Stability AI, has raised $31 million and plans to expand into video generation while potentially raising more funds to grow its business in the competitive media generation space.

For more in-depth details, read the full article.

Microsoft starts paying publishers for content surfaced by Copilot

Microsoft has announced that it will start paying publishers for the content used by its AI assistant, Copilot Daily, which provides daily summaries of weather and news. This feature will be launched in the U.S. and U.K. and will use content from authorized sources like Reuters and The Financial Times.

This decision follows a trend among AI companies to make similar agreements to avoid copyright issues and to acquire training data legally. This move comes at a time when journalism is facing job cuts and struggles to find sustainable business models due to competition for advertising revenue from big tech companies.

For more in-depth details, read the full article.

| Latest Research and Discoveries

Influence of a Large Language Model on Diagnostic Reasoning: A Randomized Clinical Vignette Study

Diagnostic errors can harm patients, and efforts to reduce these errors have had limited success. Researchers are looking at how AI models like GPT-4 can help doctors make better diagnoses.

In a study with 50 U.S. doctors, participants were split into two groups: one used GPT-4, and the other used traditional diagnostic tools. They worked on real patient cases to see how well they could diagnose.

Key Findings:

• No Major Improvement: Using GPT-4 didn’t significantly improve diagnosis accuracy. The scores were similar for both groups.
• Slight Time Savings: Doctors using GPT-4 spent a little less time on each case, but the difference wasn’t significant.
• Better AI Performance: GPT-4 performed better on its own than the doctors did with traditional methods.

Overall, the study shows that while GPT-4 can help, more research is needed to find the best ways to use AI in healthcare. Training doctors on how to work with AI is also important.

Read more: https://www.medrxiv.org/content/10.1101/2024.03.12.24303785v1

Imagine yourself: Tuning-Free Personalized Image Generation

IBM and NASA developed the Prithvi WxC, a new model designed for weather and climate forecasting. It has 2.3 billion parameters and uses data from the MERRA-2 dataset, which includes 160 atmospheric variables.

Model Features:

• Transformer-based Architecture: Efficiently processes data to capture weather patterns.
• Adaptable: Can handle various forecasting tasks.

Performance Highlights:

• It accurately predicted Hurricane Ida's path, making fewer mistakes than other models.
• It outperformed traditional methods in predicting air temperature.

Prithvi WxC marks a big step forward in weather forecasting. Its ability to perform different tasks within one model could change how we predict weather and climate, improving accuracy and reducing computing needs.

Read more: https://arxiv.org/abs/2409.13598?

| Updates From ModelsLab

?? BIG NEWS, EVERYONE! ??

We're super excited to reveal a game-changing update to ModelsLab that's going to take your creativity to the next level! Introducing... ?? ModelsLab 2.0! ???

Here's what you can expect from this amazing upgrade:

• Imagen: Transform ideas into stunning visuals instantly! ??
• Audiogen: Create lifelike audio experiences with ease! ??
• 3D Verse: Dive into immersive 3D content creation! ???
• Video Fusion: Seamlessly blend video elements for captivating storytelling! ??
• LL Master: Engage in dynamic conversations powered by advanced language models! ??

With these powerful tools at your fingertips, your creative possibilities are endless! Get ready to explore and innovate like never before! ??

Try ModelsLab!

Affiliate Program

Join our affiliate program and start earning commissions for your referrals.

Help your network learn more, build more on AI, and get paid for it. Learn more by signing up and checking out your dashboard - https://modelslab.com/

Join Our Community

Join our community on LinkedIn, Instagram, and X and connect with like-minded people who share similar interests and keep tabs on our communications. Share your stories, showcase what you have been working on, and learn from others through our Discord.

| Keep Eyes On This

Old Photo Restoration with ComfyUI

ThinkDiffusion has released a free guide on using ComfyUI for restoring old photos. This guide includes step-by-step instructions and uses ControlNet and ReActor to enhance sharpness, contrast, and color in faded images.

Learn more

Flux Latent Upscaler Workflow

Enhance image quality with ComfyUI’s Flux Latent Upscaler. This workflow uses a two-pass process to first generate a low-resolution image and then upscale it by 2x, preserving details and allowing for an optional film grain effect. It takes about 280 seconds per image on an RTX 4090.

Explore the workflow on GitHub

ComfyUI Advanced Live Portrait

A new ComfyUI extension for real-time facial expression editing and animation. You can edit expressions in photos and videos, create animations, and extract expressions from sample photos. Available via ComfyUI-Manager for easy installation.

Check it out on GitHub

ComfyUI v0.2.0 Update

ComfyUI’s latest update improves queue management, node navigation, and overall user experience. Features include new Flux ControlNets, enhanced queue management, better image display, and more.

Read the full release notes

Anifusion.AI: AI Comic and Manga Creation

Create comics and manga with Anifusion.AI’s all-in-one platform. It offers text-to-comic generation, customizable layouts, and built-in editing tools. Free and premium tiers are available.

Discover more

Skybox AI: Create 360° Worlds

Skybox AI lets you create immersive 360° panoramic worlds using AI. It supports text-to-image and sketch-to-image functionality, high-resolution outputs, and depth maps for 3D applications.

Learn more

Text-Guided Image Colorization Tool

GitHub user nick8592 has released a tool for interactive image colorization using Stable Diffusion and BLIP captioning. It allows you to specify colors for objects in grayscale images and includes a user-friendly interface.

Check it out on GitHub

ViewCrafter: Novel View Synthesis Tool

ViewCrafter generates new viewpoints from single or sparse reference images, with precise camera control and two pre-trained models. It’s open-source and designed for research purposes. Explore ViewCrafter

RB-Modulation: AI Image Personalization

RB-Modulation offers a training-free method for customizing AI image generation, enabling stylization and content-style composition from single images without unwanted content leakage.

Read more about RB-Modulation

P2P-Bridge: 3D Point Cloud Denoising

P2P-Bridge, developed by ETH Zurich and other collaborators, offers a new method for denoising 3D point clouds using Diffusion Schr?dinger bridges. It supports RGB data, DINOV2 features, and is compatible with various point clouds.

Visit the GitHub repository

HivisionIDPhotos: AI ID Photo Tool

HivisionIDPhotos generates ID photos using AI with features like lightweight portrait matting and various size options. It supports offline and cloud-based inference.

Explore HivisionIDPhotos

That’s a wrap for this edition of our AI newsletter! We hope you found the updates useful and engaging. To keep up with the latest AI news and insights, subscribe to our newsletter.

Subscribe Now to get the newest AI developments and exclusive content delivered directly to your inbox. Join our community to stay informed about the future of technology!