AI In Brief: A Comprehensive Overview
OpenAI CEO Advocates for Energy Breakthrough to Advance AI
Sam Altman, the visionary CEO of OpenAI, has emphasized the dire need for a breakthrough in energy production to fuel the development of increasingly capable and power-hungry AI models. During a panel discussion at Davos, he asserted, “There’s no way to get there without a breakthrough.” Altman’s personal investments in renewable energy sources, particularly nuclear fusion, underscore his commitment to finding a sustainable solution to this challenge. Notably, he has invested $375 million in Helion Energy, a nuclear fusion startup aiming to provide Microsoft with clean energy in the coming years.
The vast computational demands of modern AI models, often composed of billions of parameters, necessitate enormous amounts of energy during training. For instance, OpenAI’s previous-generation GPT-3 system reportedly consumed a staggering 936 megawatt hours (MWh) of energy, equivalent to the annual energy usage of approximately 90 households. As AI models continue to grow in complexity, their energy requirements are projected to escalate further. Aiden Gomez, CEO of Cohere, echoed this concern in a discussion at Davos, stating, “We’re not done with scaling [LLMs] – we still need to push up.”
Google DeepMind’s AlphaGeometry: AI System Proves Geometric Theorems at Human Level
Researchers at Google DeepMind have achieved a remarkable milestone in AI reasoning with the development of AlphaGeometry, a system that exhibits near-human capabilities in proving geometric theorems. The system, detailed in a recent Nature publication, combines a language model with a symbolic deduction engine. The language model generates potential mathematical strategies to solve a given problem, while the symbolic deduction engine attempts to derive a final solution.
To train AlphaGeometry, researchers utilized 100 million samples of synthetic data depicting random geometric diagrams. The system was tasked with learning the relationships between points and lines within these shapes to uncover geometric proofs. In a benchmarking test, AlphaGeometry successfully solved 25 out of 30 geometry questions from Olympiad competitions, achieving a performance level comparable to that of human gold medalists. Google DeepMind has released the code for the model, enabling further exploration and advancements in automated geometric reasoning.
Concerns Raised over Democratization of Healthcare by Medical AI Chatbots
The World Health Organization (WHO) has expressed reservations about the potential of medical AI systems to effectively address healthcare needs in poorer countries, particularly if these systems are developed by organizations in wealthier nations without adequate data diversity. While developers like Google envision AI as a tool to improve healthcare accessibility for underserved populations, WHO officials warn that the technology may exacerbate existing inequities and biases if it is not representative of the patient populations it is intended to serve.
Alain Labrique, the WHO’s director for digital health and innovation, emphasizes the importance of preventing the amplification of inequities through AI-driven healthcare advancements. He advocates for independent third-party audits of AI technologies before their release and calls for broader involvement beyond large tech companies in the development of these systems. Concerns have been raised regarding the potential impact of factors such as different accents, languages, and medical histories on the performance and accuracy of medical AI systems.
Amazon Introduces Experimental AI Shopping Assistant
Amazon has unveiled an experimental AI shopping assistant integrated into its mobile app, empowering consumers to engage in conversational queries about specific products. The “Looking for specific info” tab, previously featuring product reviews and answers to frequently asked questions, has been replaced with a large language model. This system ingests and summarizes information from the product listing page, providing concise responses to user inquiries.
The AI assistant is designed to focus solely on answering questions related to the product being viewed, offering convenient access to product information. However, it cannot compare products, suggest alternatives, add items to virtual carts, or disclose pricing history. An Amazon spokesperson confirmed the testing of the chatbot, highlighting its potential to improve the shopping experience by providing convenient access to product information.
The chatbot’s versatility extends beyond product-specific inquiries, reportedly capable of generating jokes, poems, and even code based on information derived from product descriptions. However, like many chatbots, it is susceptible to hallucination, and its responses should be interpreted with caution.
Call to Action:
As AI continues to make rapid strides, it is crucial to stay informed about the latest developments and implications of this transformative technology. Subscribe to our newsletter and follow us on social media to stay abreast of groundbreaking AI advancements and engage in thought-provoking discussions. Together, let’s navigate the evolving landscape of AI and shape a future where technology empowers humanity.