Biden's AI Revolution
PLUS: MIT and NVIDIA's innovative techniques for more efficient AI models, and more!
Good morning. Today is Tuesday, October 31st, and we’re covering the President Biden's recent executive order sets comprehensive AI regulations, Section 32's led by former Google executive Andy Harrison unique AI investments, and MIT and NVIDIA's AI model enhancements. New to The Intelligence Age? Sign up here.
Drop us a line anytime at [email protected] with feedback.
News & Insights
Biden Issues Sweeping Executive Order That Touches AI Risk, Deepfakes, Privacy
President Biden's recent executive order, the first of its kind, delineates comprehensive regulations for generative AI systems. The order touches upon multiple facets of AI usage, including risk mitigation, deepfakes, and privacy concerns. It calls for mandatory testing of advanced AI models to prevent misuse for weapon creation, and suggests watermarking AI-generated media. Furthermore, it addresses potential job displacement due to AI and privacy concerns.
A distinctive approach is highlighted in the order: Biden will leverage the federal government's purchasing power to enforce AI standards. Federal agencies will only ink contracts with companies adhering to the newly outlined AI regulations. This move aims to drive compliance, using the government's influence as a significant client. arstechnica
Why does this matter?
While this order primarily targets the US, it could have ripple effects globally. As multinational companies strive to align with these regulations, we might see an international standardization of AI ethics and practices, accelerating the responsible adoption of AI worldwide.
How Former Googlers’ VC Firm Invests In Everything From LLMs To AI Doing Drug Discovery
Section 32, led by former Google executive Andy Harrison, is making significant strides in AI investment. With a keen eye for second- and third-order problems, this VC firm is betting on areas untouched by cloud players. They're funding AI in infrastructure, enterprise services, cybersecurity, precision medicine, and computational biology. Their portfolio includes Cohere, Inceptive, BigHat Biosciences, Inworld AI, and Metaphysic, all demonstrating unique applications of AI.
AI's continued evolution, coupled with cheaper storage and processing power, has enabled massive LLMs that capture our imagination. Yet, Harrison reminds us, it's just software. Section 32 seeks to invest in AI companies that drive real ROI. As Harrison puts it, "It's expensive software because of the compute and because of the talent required. So it has to be that much better... Or it has to do something we couldn't previously do with software in order to drive high margins." crunchbase
Why does this matter?
Harrison's perspective raises a compelling question: What happens when AI becomes a commodity and loses its 'wow' factor? As AI continues to permeate every industry, its fundamental value may shift from novelty to utility, and the focus will be less on the technology itself and more on the real, tangible benefits it delivers.
Together with The Artificially Intelligent Enterprise
Navigating the Future of Work
The Artificially Intelligent Enterprise, is a weekly newsletter curated by Mark Hinkle, an emerging technology executive with a proven track record of bringing new technologies to market.
Gain insights into new developments in Generative AI and how they're influencing industries and workplaces.
Receive tips for supercharging your productivity by solving everyday challenges through AI.
Stay ahead with curated links to cutting-edge productivity apps that leverage Generative AI.
The Artificially Intelligent Enterprise isn't just another tech newsletter; it's your roadmap for navigating the future of work in an AI-driven world.
Please support our sponsors!
Aragon: Professional headshots generated instantly through A.I. technology. (website) (sponsored)
Pictorial: A service where a Generative AI is used to effortlessly create graphics for web applications. (website)
CodeComplete: Enterprise AI assistant for coding with self-hosting and customization. (website)
Accelerating AI Tasks While Preserving Data Security
SecureLoop, a new tool developed by MIT researchers, is set to revolutionize data security in AI applications by optimizing the design of deep neural network accelerators. This development arises from the ongoing struggle of balancing high-speed computations with the need of safeguarding sensitive data. SecureLoop shows promise in meeting these dual goals, potentially transforming the way data security is approached in AI.
SecureLoop cleverly circumnavigates the tricky terrain of data encryption and authentication, which commonly slow down computations and increase energy usage. By introducing a method that minimizes the frequency of off-chip memory access, SecureLoop effectively reduces the computational cost, yielding a design that is secure, swift, and energy-efficient. MIT News
Around The World
- UK government initiates £100M AI fund aimed at aiding the treatment of incurable diseases.
Excel and ChatGPT are the best combo.
You can automate your data processing easily.
I show you how to use AI with Excel without typing formulas:
— Paul Couvert (@itsPaulAi)
Oct 30, 2023
New Techniques Efficiently Accelerate Sparse Tensors For Massive AI Models
Harnessing zeros for a tech leap. MIT and NVIDIA researchers have finessed two techniques to improve processing of sparse tensors, the data structures that power generative artificial intelligence. Exploiting the presence of zero values in these tensors, the scientists could significantly enhance both performance and energy efficiency of AI models.
Firstly, they've developed a method that expands the horizon of sparsity patterns, allowing the hardware to efficiently locate non-zeros. This approach, coined 'hierarchical structured sparsity', breaks down tensor values into smaller blocks, each with its unique yet simple sparsity pattern. By continually combining blocks into larger levels, the technique maintains simplicity, enabling efficient skipping of zeros and reducing redundant computation.
Tailors and Swiftiles: Overbooking, the AI way. The second technique addresses storage constraints. Just as airlines overbook flights, the researchers have 'overbooked' storage space, fitting larger chunks of data (tiles) into the on-chip memory buffer by banking on a high proportion of zeros. If a tile has more non-zeros than can fit, only the extra data are re-fetched, thanks to a method named 'Tailors'. Simultaneously, 'Swiftiles' rapidly estimates the ideal tile size for overbooking, reducing computational demands. futurism
Written by Isaac R. Ward, Casey Clifton, and Alex Brogan.
Send us feedback at [email protected] and help us provide the best coverage of Artificial Intelligence possible.
Interested in reaching smart business leaders like you? To become an Intelligence Age partner, apply here.