AI Weekly Rundown (May 25 to May 31)
Major AI announcements from Microsoft, Google, OpenAI, Mistral AI, and more.
Hello Engineering Leaders and AI Enthusiasts!
Another eventful week in the AI realm. Lots of big news from huge enterprises.
In today’s edition:
🕵️ Microsoft's Recall AI: Efficiency upgrade or privacy nightmare?
🎧 AI-powered headphones let you listen to one person in a crowd
⚡ xAI’s supercomputer that’s 4x larger than existing GPU clusters
🚨 Google AI Overview gives dangerous and wrong answers
🚀 GPT-4 is a better financial analyst than humans
🎨 Canva redesigned its platform with new AI features
🔒 OpenAI forms an AI safety committee
😲 Shocking details about Sam Altman's firing last year
🔄 Ex-OpenAI safety lead joins rival Anthropic to lead a new team
🤖 Mistral AI introduces Codestral, a code-generation AI model
📰 OpenAI licenses content from Vox Media and The Atlantic
💻 Google infuses Chromebook Plus with powerful AI features
⚠️ OpenAI reports misuse of its AI for "deceptive activity"
🏫 OpenAI introduces ChatGPT Edu, an affordable option for universities
🤝 Tech giants form an industry group to develop next-gen AI chip
Let’s go!
Microsoft's Recall AI: Efficiency upgrade or privacy nightmare?
Microsoft has recently released Recall, a controversial AI-powered tool that logs everything you see and do on your computer, including app usage, live meeting communications, websites visited, and more. By simply performing a "Recall" action, users can retrieve any information they've interacted with on their device, presented in the context of a specific time period.
Microsoft assures users that the Recall index remains local and private on-device. Users can pause, stop, or delete captured content and can choose to exclude specific apps or websites. And Recall automatically excludes InPrivate web browsing sessions in Microsoft Edge and DRM-protected content.
Currently, Recall is exclusively compatible with new "Copilot Plus PCs" equipped with Qualcomm's Snapdragon X Elite chips, which feature the necessary neural processing unit (NPU).
AI-powered headphones let you listen to one person in a crowd
Researchers at the University of Washington have developed a groundbreaking artificial intelligence system called "Target Speech Hearing" that allows headphone wearers to isolate and listen to a single speaker in a noisy environment. By simply looking at the desired speaker for three to five seconds, the user can "enroll" them, and the system will cancel out all other sounds, playing only the enrolled speaker's voice in real-time, even as the listener moves around.
The proof-of-concept device, which uses off-the-shelf headphones fitted with microphones and an on-board embedded computer, builds upon the team's previous "semantic hearing" research. The system's ability to focus on the enrolled voice improves as the speaker continues talking, providing more training data. While currently limited to enrolling one speaker at a time and requiring a clear line of sight, the researchers are working to expand the system to earbuds and hearing aids in the future.
xAI’s supercomputer that’s 4x larger than existing GPU clusters
Elon Musk has recently disclosed plans to build a groundbreaking supercomputer to power the next iteration of xAI’s chatbot, Grok. According to a presentation made to investors in May, as reported by The Information, Musk aims to have this computing system operational by the fall of 2025. The ambitious project may involve a collaboration between xAI and Oracle to develop this massive computing infrastructure.
The proposed supercomputer will consist of interconnected clusters of Nvidia's state-of-the-art H100 graphics processing units (GPUs). Musk revealed that upon completion, the scale of this system will surpass the largest existing GPU clusters by at least fourfold!
Google AI Overview gives dangerous and wrong answers
Google's AI Overviews feature, which generates AI-powered responses to user queries, has been providing incorrect and sometimes bizarre answers. From suggesting glue on pizza, staring at sun for health benefits to claiming that former US President Barack Obama is Muslim, the feature has left users questioning the reliability of AI-generated search results.
While Google maintains that these mistakes result from uncommon queries and are being used to refine the product, the technology's widespread deployment has highlighted the challenges and risks of integrating AI into search engines.
GPT-4 is a better financial analysts than humans
In a groundbreaking study, researchers at the University of Chicago have discovered that GPT-4, a large language model, can analyze financial statements and predict future earnings direction with remarkable accuracy. Even without narrative context, GPT-4 outperforms human financial analysts and achieves prediction accuracy on par with state-of-the-art machine learning models.
The AI model's success is not simply a result of its training memory but rather its ability to generate valuable narrative insights about a company's future performance. Notably, trading strategies based on GPT-4's predictions yield higher returns and risk-adjusted performance than those based on other models, especially for small companies.
Canva redesigned it’s platform with new AI features
Canva is launching a redesigned platform with new AI features and tools for professional teams and workspaces. The company has introduced Canva Enterprise, a specialized tier offering more control over collaboration, brand management, and security for larger organizations.
Ad Tech integrations with Google, Meta, and Amazon streamline the ad creation process within Canva. Data autofill automates the creation of data-driven designs by integrating with sources like Salesforce and MLS.
New features include AI style matching for brands, customizable folder displays, and the ability to "star" designs and templates for easier access. Canva Docs now has a suggestion mode for editors and colorful highlight blocks for text emphasis. New "Magic Studio" AI tools include:
Automatic clip highlighting.
Background noise reduction for video editing.
A text-to-graphic image generator.
OpenAI forms an AI safety committee
OpenAI Board formed a Safety and Security Committee led by directors Bret Taylor, Adam D’Angelo, Nicole Seligman, and Sam Altman. Over the next 90 days, the committee will evaluate and further develop OpenAI’s processes and safeguards. Then, the committee will share its recommendations with the full Board.
Following the full Board’s review, OpenAI will publicly share an update on adopted recommendations. OpenAI has recently begun training its next frontier model, and the resulting systems from the Safety and Security team’s recommendation may bring it safely to the next level of capabilities on its path to AGI.
Shocking details about Sam Altman’s firing last year
Former OpenAI board member Helen Toner revealed shocking details about Sam Altman’s firing in November 2023. The board had found it difficult to trust Altman after he failed to disclose he owned the OpenAI Startup Fund, gave inaccurate information about the company's safety processes, and tried to push Toner off the board after she published a critical research paper. She claims the board found out about the launch of ChatGPT on Twitter, not from Altman.
Toner also pointed to Altman's history of being fired from previous jobs for "deceptive and chaotic behavior." (Find the full TED AI episode linked below.)
After Altman was initially fired, there was a lot of backlash from OpenAI employees who wanted him reinstated. A week later, he was reinstated as the CEO and a new board was formed.
Ex-OpenAI safety lead joins rival Anthropic to lead a new team
Jan Leike, a former co-head of the OpenAI safety team, has joined Anthropic to lead a new "superalignment" team. Leike left OpenAI earlier this month after the team he co-led, which focused on long-term AI safety risks, was dissolved.
At Anthropic, Leike's new team will focus on various aspects of AI safety and security, including "scalable oversight, weak-to-strong generalization, and automated alignment research”. This is similar to the mission of OpenAI's former Superalignment team, which Leike previously led.
Enjoying the weekly updates?
Refer your pals to subscribe to our newsletter and get exclusive access to 400+ game-changing AI tools.
When you use the referral link above or the “Share” button on any post, you'll get the credit for any new subscribers. All you need to do is send the link via text or email or share it on social media with friends.
Mistral AI introduces Codestral, a code AI model
Codestral is a 22B parameter generative AI model designed specifically for code generation tasks. It is trained in over 80 programming languages, including popular ones like Python, Java, C++, and JavaScript. It excels at code completion, writing tests, filling in partial code, and testing code, thereby improving developer productivity and reducing errors.
The model sets a new performance/latency standard for code generation tasks compared to previous models. It can be downloaded on HuggingFace and is available on Mistral’s API platform through instruct and fill-in-the-middle endpoints. It can be easily integrated into VScode plugins and used for free on Le Chat.
OpenAI licenses content from Vox Media and The Atlantic
OpenAI has formed content and product partnerships with Vox Media and The Atlantic. OpenAI will license content from these media powerhouses for inclusion in the chatbot’s responses.
In turn, The Atlantic’s product team will have privileged access to OpenAI tech, give feedback, and share use cases to shape and improve future news experiences in ChatGPT and other OpenAI products. Vox and OpenAI will also collaborate using OpenAI’s technology to develop innovative products for Vox Media’s consumers and advertising partners.
Google infuses Chromebook Plus with powerful AI features
Google’s Chromebook Plus has new built-in Google AI and gaming features, Help me write, Chat with Gemini, Magic Editor, and more.
“Help me write” allows users to get AI-generated text suggestions, tone changes, rewriting, and more right within their writing apps on the Chromebook.
Generative AI wallpaper and video call backgrounds are built into the OS for customization, no matter what video conferencing app you’re using.
Magic Editor on Google Photos for laptops exclusively on Chromebook Plus will reimagine your photos with a few easy clicks.
Chat with Gemini on the home screen helps plan, write, learn, and more.
Chromebook Plus laptops are starting at $350 USD. Google is also launching new features that integrate more of Google across all Chromebooks to help users work better and get things done faster.
OpenAI reports misuse of its AI for 'deceptive activity'
OpenAI identified and disrupted five covert influence operations that used its generative AI models for "deceptive activity" across the internet.
OpenAI said the threat actors used its AI models to generate short comments, longer articles in various languages, made-up names, and bios for social media accounts over the last three months. These campaigns included threat actors from Russia, China, Iran, and Israel. It focused on issues including Russia's invasion of Ukraine, the conflict in Gaza, the Indian elections, and politics in Europe and the United States, among others.
OpenAI has also published a trend analysis report that describes the behavior of these malicious actors in detail.
OpenAI introduces ChatGPT Edu for universities
OpenAI announced ChatGPT Edu, a ChatGPT version custom built for universities to responsibly deploy AI to students, faculty, researchers, and campus operations. It is powered by GPT-4o and can reason across text and vision. It also includes:
Enterprise-level security and controls affordable for educational institutions
Improved language capabilities across quality and speed supporting over 50 languages supported
Advanced capabilities like data analytics, web browsing, and document summarization
Ability to build GPTs, custom versions of ChatGPT, and share them within university workspaces
Significantly higher message limits
Conversations and data are not used to train OpenAI models
Tech giants form industry group to develop next-gen AI chip
Intel, Google, Microsoft, Meta, and other tech heavyweights are establishing a new industry group called the Ultra Accelerator Link (UALink) Promoter Group to guide the development of the components that link together AI accelerator chips in data centers. The group also counts AMD, Hewlett Packard Enterprise, Broadcom, and Cisco among its members.
It is proposing a new industry standard to connect the AI accelerator chips found within a growing number of servers. It’s version one, UALink 1.0, will connect up to 1,024 AI accelerators — GPUs only — across a single computing “pod.” It is set to arrive in Q4 2024.
That's all for now!
Subscribe to The AI Edge and gain exclusive access to content enjoyed by professionals from Moody’s, Vonage, Voya, WEHI, Cox, INSEAD, and other esteemed organizations.
Thanks for reading, and see you on Monday. 😊