Nvidia stealthily drops new model

Read time: under 4 minutes

A new model reaches the top of leaderboards out of nowhere. We unpack what’s going on. And another model makes it easier to spot issues in the brain.

Today’s Insights

• Today in AI: A new model, experimental apps, and AI-powered MRIs

• Tutorial: How to edit parts of images with Krea AI

• Mistral unveils two models that run on phones

• Everything else you should know today

• 5 new AI tools to boost your productivity

• AI-Generated Images: Music Movement

TODAY IN AI

1. Good things come in pairs: First up, a mysterious new model, Llama-3.1-Nemotron-70B, just jumped to the top of the charts. Nvidia apparently gave Meta’s open-source Llama a boost, and early users say it outperforms even GPT-4o and Claude Sonnet 3.5 in several major benchmarks. It’s open source, so anyone can tweak as they wish via platforms like HuggingFace. We’ll dig into another major release — this time from Mistral — later in the newsletter.

2. Adobe engineers show off their own experimental AI projects: A text-to-animation generator. A model that lets you “rotate” 2D objects. An app that lets you turn your own voice into AI-generated sound effects. That’s just a small taste of the experimental projects shown at “Sneaks” — a biannual Adobe showcase where developers give a glimpse of the cutting-edge projects they’re working on.

3. Brain scans made easy: Researchers at UCSF have come up with an AI model that boosts the quality of 3T MRI images to look like higher-resolution 7T MRIs. This could make it easier to spot brain issues in conditions like traumatic brain injury and multiple sclerosis. The breakthrough could help more people access high-quality brain scans without needing specialized machines, although it still needs more clinical testing before it can be used widely.

PRESENTED BY SUBSTRATA

How to Use Social Signal Intelligence to Win B2B Deals 5X Faster

Complex deals are often affected by office politics, hierarchies, egos, emotions, attitudes, and power dynamics. Using Substrata can significantly increase the chances of winning B2B deals across many industries and categories.

Substrata has developed dozens of specialized AI models to help dealmakers navigate the subtle "social dance" between sellers and prospects:

• Analyze any prospect's LinkedIn profile to extract their natural personality structure and tendencies

• Monitor your prospects' fluctuating levels of interest (Signals)

• Analyze any thread or message to determine real sentiment and relative balance of power

• Simulate message drafts before sending them to eliminate potential risks or issues

• Optimize responses, follow-ups, and more

• Track your and your team's ongoing performance

Substrata uses Social Signal Processing (SSP), which is a specialized AI domain that deals with the detection, analysis, and synthesis of behavioral data from digitally mediated human-human interactions, such as B2B deal-making, sales, and HR interviews. It encompasses both verbal and nonverbal data.

For more information, visit Substrata.

THE AI ACADEMY

How to edit parts of images with Krea AI

• Go to Krea AI's website and sign up to get credits.

• On the main page, click on Generate > Realtime.

• Select Flux/HD as your model to get the best results.

• Upload your reference image and expand it to fit the canvas on the left side.

• Select Paint Brush from the menu bar and draw a rough sketch of your editing on the left side.

• Write your prompt in the prompt box, and be as accurate as possible.

• Adjust your AI strength (lower strength means more similar to the original image and vice versa)

• Once done, you can save your edited image to your gallery.

FROM THE FRONTIER

Mistral becomes ‘Ministral’ with smaller, faster models

Mistral just hit us with a one-two punch. A little over a year ago, it released the small but mighty 7B model. Now, the Paris-based startup is back with one that’s even smaller…and even mightier. Despite being half the size, Ministral 3B outperforms 7B across most tests. Meanwhile, its big brother, Ministral 8B, is reportedly now the top-performing model of its size, beating out state-of-the-art rivals from Meta and Alphabet.

Why it matters: These new models are so small they can fit on phones and computers. At the same time, they’re powerful enough to operate real-time translation apps, AR experiences, and even humanoid robots. Not to mention, since they run locally, they can be used while on the go, they have less latency, and there’s more privacy built-in compared to models that need a cloud connection.

AI & TECH NEWS

Everything else you need to know today

🤖 Helping Hand: Boston Dynamics is partnering with Toyota Research Institute (TRI) and their expertise in large behavioral models to bring advanced AI and robotic intelligence to the Atlas humanoid robot.

🌏️ Legendary Languages: Alibaba's international arm launched an updated AI-powered translation tool that supports 15 languages. The company says it even outperforms similar tools.

👋 AWS Announcements: After 15 years at AWS, Matt Wood, VP of AI, announced his departure from the company.

📗 Reading Reinvented: Amazon announced a brand new Kindle family, including a reimagined Kindle Scribe – with in-book writing and an AI-powered notebook.

PRODUCTIVITY

5 AI Tools to Supercharge Your Productivity

✅ Openmart: Get highly targeted local business leads using AI.

✅ Pitchleague: Get feedback on your pitch deck.

✅ Synthesia: Create videos with AI avatars and voiceovers in 130+ languages.

✅ Slogan Generator: Compress your marketing message into 1 sentence.

✅ Sendbird: Create an AI chatbot for your websites in minutes.

☁️ Join us for Startup School: Gen AI*, back by popular demand October 29, 2024 - November 21, 2024. You can learn the fundamentals of generative AI and how to apply the latest AI solutions to your business with this immersive, hands-on training series, led by experts and partners.

^{* indicates a promoted tool, if any}

PROMPT OF THE DAY

Midjourney Help

Prompt: I need your help with customizing an AI art piece generated by Midjourney for my design project. Guide me on how to modify and customize it to fit my specific design needs. Some specific information about my project includes <insert relevant details here, such as the AI art’s intended use, the design’s desired aesthetic or style, and the AI art’s specific elements to be modified or customized>

_{Source: Upbase}

AI-GENERATED IMAGES

Tune in

Midjourney Prompt: editorial photo a person 35-year-old, listens to music on headphones, motion blur effect, GoPro HERO10 Black, film grai, Kodak Portra 400
--ar 4:5 --style raw --v 6.1

Acquire new customers and drive revenue by partnering with us

Superhuman is the world’s biggest AI newsletter for businesses and professionals with 800,000+ readers and 1.5 Million followers on socials working at the world’s leading startups and enterprises. Companies like Amazon, Hubspot, and Salesforce feature their products in Superhuman. You can learn more about partnering with us here.

🧞Your wish is my command

Got more feedback or just want to get in touch? Reply to this email and we’ll get back to you.

Thanks for reading.

Until next time!

Zain & the Superhuman AI team