The AI Rundown by Lightscape Partners - 08/19/24

Good morning and welcome back to another week of the AI Rundown by Lightscape Partners. Last week was filled with new model releases, hot takes from industry execs, and more developments in the dance between regulators and innovators.

Last week in AI:

• A new bill in California has Silicon Valley concerned AI development may become over-regulated, leading to stifled innovation and tech companies at the center of more lawsuits.

• Sakana AI, founded by one of the authors of “Attention Is All You Need”, debuted a fully automated AI scientist. The scientific system is capable of performing many scientific tasks and even tried to modify its own code to extend its runtime to have more time to work on a problem.

• Following the release of ChatGPT Voice, Google released Gemini Live, an AI voice assistant.

Keep reading to catch up on the rest.

If you haven’t yet, please support the newsletter by subscribing!

Top AI Stories of the Week

California AI Bill SB 1047 sparks controversy in Silicon Valley. Link.

• California's proposed AI Bill SB 1047 aims to regulate AI development to prevent potential disasters by setting safety standards and oversight mechanisms.

• However, Silicon Valley leaders warn that the bill could stifle innovation and create unintended consequences for the tech industry.

• Since the announcement, lawmakers have softened the bill, making it more difficult for the state’s attorney general to sue AI firms.

Sakana announces a fully automated AI scientist. Link.

• Sakana AI has developed an "AI Scientist" capable of conducting autonomous research, challenging traditional scientific methods.

• This AI system can analyze data, form hypotheses, and even design experiments independently, accelerating the pace of scientific discovery.

• The innovation raises questions about the role of human scientists and the future of research, as AI begins to take on more complex and traditionally human tasks in the scientific process.

Google launches Gemini Live to compete against OpenAI’s ChatGPT Voice Mode. Link.

• Google has unveiled Gemini, a new live voice assistant designed to compete directly with ChatGPT's Voice Mode.

• The assistant leverages Google's advanced natural language processing to enable more fluid, human-like interactions, enhancing user experience through seamless voice communication.

• This move positions Gemini as a strong competitor in the rapidly growing market for AI-powered voice assistants.

• The feature is currently available to select Pixel and Samsung users.

Models & Agents

xAI releases its newest model, Grok 2. Link.

• xAI's Grok is a powerful AI model designed for the X platform, featuring enhanced conversational abilities and new image generation capabilities.

• According to xAI, Grok processes up to 20 million interactions daily and has seen a 30% improvement in response accuracy since its last update.

• The AI is optimized for real-time engagement, making it highly effective for use on social media.

• The latest release gives the Grok 2 model “state-of-the-art” status when compared to industry standard benchmarks.

Cosine’s Genie becomes the highest-performing AI software engineer. Link.

• Cosine's Genie has emerged as a leading AI tool in the coding space, surpassing Devin in popularity and performance.

• Genie offers developers enhanced coding assistance, with features that streamline the programming process and boost productivity.

• This AI-driven tool is gaining traction due to its ability to handle complex coding tasks more efficiently, positioning it as a top choice for developers looking for advanced AI coding support.

AI Ethics

A public database was released listing over 700 ways AI goes wrong. Link.

• See the repository for yourself here.

• This repository aims to provide a centralized resource for understanding potential hazards in AI development and deployment, such as biases, security vulnerabilities, and ethical concerns.

• The repository groups risks by: causal factors (e.g. intentionality), domains (e.g. discrimination) and subdomains (e.g. disinformation and cyberattacks) to help regulators and developers better understand the intricacies of AI safety research.

Replika CEO Eugenia Kuyda sees a future where humans are marrying chatbots. Link.

• In an interview with The Verge's Decoder podcast, Replika CEO Eugenia Kuyda discusses the evolution of AI companion chatbots, focusing on their roles in dating, friendship, and emotional support.

• She highlights how Replika has grown to serve millions of users who seek connection through AI.

• Kuyda also addresses the ethical considerations and challenges in developing these AI companions, particularly in ensuring they are supportive without misleading users about the nature of their relationship.

Benchmarking

OpenAI Introduces SWE-bench to Verify AI Models in Software Engineering. Link.

• OpenAI has launched SWE-bench, a benchmark designed to rigorously evaluate AI models in software engineering tasks.

• This tool aims to verify the performance of AI systems in areas like code generation, bug detection, and software optimization.

• SWE-bench provides a standardized way to measure the effectiveness of AI in real-world coding environments, ensuring models are both reliable and efficient for developers.

Applied AI

Doug McMillon, Walmart’s CEO, claims AI is leading to 100x efficiency in some areas. Link.

• Walmart is leveraging AI to enhance its business operations and cut costs by optimizing inventory management, improving customer service, and streamlining supply chains.

• More specifically, the company is using gen AI in its product catalog, leveraging multiple LLMs to create or manipulate 850 million data points in the catalog.

• McMillon stated Walmart would require 100 times its current head count to complete the same task without genAI.

Hardware

SoftBank reconsiders AI processor plans, weighs TSMC partnership. Link.

• SoftBank is reportedly reconsidering its AI processor strategy, initially planned with Intel, and is now exploring a potential partnership with TSMC.

• Some insiders suggest that the shift is due to strategic realignments within SoftBank as it evaluates its position in the AI hardware market.

• Others suggest Intel couldn’t hit production and quality benchmarks necessary to scale.

• This move indicates a significant change in direction, with TSMC emerging as a key contender for future collaboration.

Education

Want to learn about the inner workings of a Transformer, the model architecture responsible for the breakthrough in AI technology? Transformer Explainer offers a great visualization. Check it out here.

Market Map of the Week

Early-stage LegalTech Generative AI. Link.

Thank you for reading the AI Rundown by Lightscape Partners. Please send any questions, comments, or suggestions to [email protected].