Apple's AI comeback

ALSO: AI's voracious energy demands

Read time: under 4 minutes

Welcome back, Superhuman

Speculation has been rife in recent months that Apple might have missed the AI train. But a number of big moves in the past week — including a new AI model and reports of a blockbuster deal with Google — hint at something major cooking behind the scenes.

First time reading? Sign up to get this newsletter here

TODAY’S MENU

  • Apple makes big AI moves with new model (and more)

  • How to use GPT-4’s data analysis and PDF features for free

  • AI’s energy demands raise concerns among experts

  • Chart: Types of AI systems through the decades

  • 5 new AI tools to boost your productivity

  • AI Generated Images: Musical instruments are everywhere

NEWS

Today in AI & Tech

Source: Under Armour advertisement, screenshot

  • Editing Ethics: Sportswear company Under Armour sparked controversy online after it revealed that an Instagram ad featuring boxer Anthony Joshua had been created with the help of AI.

  • Silicon Siesta: A new Bernie Sanders-led bill cited advancements in AI and automation as reasons why Congress should cut the federal workweek to 32 hours.

  • Robo-Advisor: Morgan Stanley has named a top executive as the bank’s first head of AI, illustrating the financial world’s growing interest in generative LLMs.

  • AI Apprentices: Mercedes-Benz announced it’s partnering with Apptronik on a line of humanoid robots that would help with low-skill, physically demanding labor during the manufacturing process.

  • Promise Kept: Staying true to his word, Elon Musk (and xAI) open sourced the code of its Grok-1 AI model.

NEXT IN AI

Apple is making big AI moves with a brand new AI model (and more)

Screenshot from Apple’s MM1 paper. Source: Brandon McKinzie/Apple

Just as the tech pundits were busy writing Apple's AI obituary, the company has come out of the shadows to announce its own AI model that’s capable of rubbing shoulders with Google’s Gemini and OpenAI’s GPT-4 models.

The company quietly published a new paper about MM1 – a family of multimodal large language models (MLLMs). It details the breakthroughs its researchers have made in multimodal AI – combining text and image data for training large language models.

What to know: 

  • This release is impressively transparent for a company known for its secrecy

  • MM1’s largest model can support 30 billion parameters and can perform “in-context predictions, multi-image and chain-of-thought reasoning” thanks to its large-scale multimodal pre-training

  • The researchers say their work has enabled models to achieve “state-of-the-art few-shot results across multiple benchmarks.”

But that may only be the tip of the iceberg. According to a report from Bloomberg released earlier today, Apple is in talks to let Google's Gemini power future iPhone AI features. Google’s Gemini models already power a number of features in Samsung smartphones.

What new products can we expect? Apple is already using generative AI for internal processes and customer service. But it’s expected to announce a suite of new AI-related features during its worldwide developers conference in June. 

Siri could see a major upgrade for iOS 18, including improvements to how it understands and responds to human language. Apple users can also expect AI-created playlists in Apple Music and AI-powered productivity tools for Pages and Keynote.

Reading between the lines: Apple is reportedly investing about $1 billion a year into LLMs. The company also reportedly purchased generative AI startup DarwinAI last week, which specializes in making LLMs faster and more efficient. Coupled with its new model and a potential partnership with Google, Apple could be on the verge of making a major AI announcement.

PRESENTED BY INNOVATING WITH AI 

They say AI will simplify your life – but the endless lists of 1,001 ChatGPT Ideas make things more complicated, not less.

That’s why Innovating with AI, led by a 20-year tech veteran who’s worked for Harvard, MIT and The World Bank, created The Free No-Code AI Toolkit, a carefully curated guide to the most powerful tools and best use cases for AI in your office workflow. They’ve combed through all the options, cut out the distractions, and applied these automations to our own $1MM/year tech business – and now they’re ready for you, free and with zero coding knowledge required.

AI AT WORK

How to use GPT-4’s data analysis and PDF features for free

Sample document summarized with Copilot

If you’re on the free version of ChatGPT, you’re not able to access several features like PDF uploads, data analysis, chart creation, and more. But Microsoft’s copilot — which uses GPT-4 — lets you access these features at no cost. Here’s how:

  • Go to the Copilot website

  • Log in or sign up for a new account

  • In the chat window you’ll see a pin icon to attach a file, click the button and upload a PDF or spreadsheet

  • Ask Copilot to analyze your document or produce a specific chart based on the data

In our tests we found the PDF upload feature to be more reliable than the data upload and visualization feature, which can occasionally misfire. We recommend removing any sensitive company or personal information before uploading files to Copilot.

AI & ENERGY

How AI is transforming the power grid

The average American home consumes 29 kilowatt-hours (kWh) of electricity every 24 hours. But ChatGPT is on a whole other level: It uses 500,000 kWh each day, according to the New Yorker. Not to mention the colossal amount of processing power it took to create ChatGPT — along with its growing list of rivals. 

Energy experts fear we’re running up against a problem: The world isn’t prepared for the intense energy demands that AI requires. 

What’s the scale of the problem? 

  • AI systems are housed inside data centers, which use power in a variety of ways, including through energy-intensive cooling equipment.

  • There are nearly 6,000 data centers around the world.

  • The amount of electricity data centers use is expected to triple by 2030. At that point, data centers could represent about 7.5% of America’s energy usage. 

That may not sound like a lot, but even a modest increase can overwhelm the globe’s already strained energy system.

What can we do about it? AI companies say they need more power — and fast. Both OpenAI and Microsoft have petitioned for more nuclear energy initiatives. Others are trying to find ways to make AI computing more efficient and sustainable.

It’s not all bad: Although data centers contribute to the global emissions problem, they might also be part of the solution. Sustainability experts think AI could have numerous practical applications, like helping to boost the effectiveness of solar, wind, and other alternative energies.

CHART

Types of AI systems through the decades

EpochAI/Our World In Data

An analysis by EpochAI shows a timeline of notable AI models over the years, categorized by type. Here’s a breakdown of what each category means.

  • Other: 3D reconstruction, autonomous driving, video and audio processing, and search algorithms

  • Speech: Voice assistants and other models that interpret and generate spoken language

  • Games: Models that learn how to play games against skilled opponents

  • Vision: Photo tagging systems and image recognition software

  • Multimodal: Systems that integrate text, images, and audio (ie ChatGPT)

  • Biology: Models created for scientific purposes

  • Image Generation: Systems that turn textual prompts into images (ie Midjourney and DALL-E)

  • Language: Models that understand and translate language (ie Google Translate)

PRODUCTIVITY

5 AI Tools to Supercharge Your Productivity

MyMemo: Gather articles, links, screenshots, and videos into a single, accessible platform, then ask questions about the content you’ve collected.

DryRun Security: An AI-powered tool that allows developers to receive security alerts as they write code — without needing to be security experts themselves.

Notion AI: Search, create, and ideate in Notion with AI.

Airtrain: Query and compare multiple LLMs simultaneously to quickly determine which ones are best at specific tasks.

Hypotenuse AI: Write SEO-optimized blog posts, product descriptions, social media copy, and more with AI.

* indicates a promoted tool, if any

AI GENERATED IMAGES

American Orchestra

Art/Prompts by @ifonly.ai

Acquire new customers and drive revenue by partnering with us

Superhuman is the world’s biggest AI newsletter for businesses and professionals with 600,000+ readers working at the world’s leading startups and enterprises. Companies like Amazon, Calendly, and Notion feature their products in Superhuman. Main ads are typically sold out 4 weeks in advance. You can book future ad spots here.  

🧞 Your wish is my command

What did you think of today's email?

Your feedback helps me create better emails for you!

Login or Subscribe to participate in polls.

Reviews of the day

Thanks for reading.

Until next time!

p.s. if you liked this newsletter, share it with your friends and colleagues here. For questions and feedback, email us at [email protected]. We always listen and respond