AI in Brief: Energy Breakthroughs, AI Reasoning Advancements, Medical AI Concerns, and Amazon’s AI Shopping Assistant
1. OpenAI CEO Calls for Energy Breakthrough to Support AI’s Advancements
Sam Altman, CEO of OpenAI, emphasized the need for a breakthrough in energy production to enable the continued development of increasingly capable and energy-intensive AI models during a panel discussion at Davos.
Altman voiced his preference for renewable energy sources, particularly nuclear fusion, and his commitment to investing in this technology. He personally invested $375 million in Helion Energy, a nuclear fusion startup that recently signed a deal to supply energy to Microsoft.
Training AI models composed of billions of parameters requires substantial amounts of energy. For instance, OpenAI’s previous model, GPT-3, consumed 936 megawatt hours (MWh) of energy, equivalent to the annual energy consumption of approximately 90 households.
In the pursuit of larger models, Aiden Gomez, CEO of Cohere, acknowledged the ongoing need to scale up LLMs (Large Language Models) during a separate discussion at Davos.
2. Google DeepMind’s AlphaGeometry: A Breakthrough in AI Reasoning
Researchers at Google DeepMind have made significant strides in developing an AI system, AlphaGeometry, capable of solving geometric theorems at a level comparable to human mathematics Olympiad gold medalists.
Published in Nature last week, AlphaGeometry comprises a language model and a symbolic deduction engine. The language model generates potential strategies to solve a given problem, while the symbolic deduction engine attempts to derive a final solution.
Co-authors Trieu Trinh and Thang Luong stated, “With AlphaGeometry, we demonstrate AI’s growing ability to reason logically and to discover and verify new knowledge.”
Trained on 100 million samples of synthetic data, AlphaGeometry learned the relationships between points and lines in geometric shapes to derive proofs.
In a benchmark test, AlphaGeometry solved 25 out of 30 geometry questions from Olympiad competitions within a few hours, a performance comparable to the average human gold medalist.
Google DeepMind has released the code for AlphaGeometry, enabling further exploration and research in this field.
3. WHO Raises Concerns about Medical AI Chatbots in Healthcare Democratization
The World Health Organization (WHO) has expressed concerns that medical AI systems developed by organizations in wealthier nations may not adequately address the needs of poorer countries if they are not trained on diverse data.
Tech giants like Google believe that AI can improve healthcare access for underserved populations. However, officials at the WHO caution that these technologies may not provide optimal service if they are not designed with diverse patient data representation.
Alain Labrique, WHO’s director for digital health and innovation, emphasized the importance of preventing the propagation of inequities and biases through AI-driven healthcare.
Labrique and his colleagues advocate for a balanced approach, urging against the dominance of large tech companies in medical AI development and calling for independent audits of these technologies before their release.
Current efforts in medical AI include the development of models capable of generating clinical notes, aiding doctors in diagnosing diseases, and more.
However, challenges such as diverse accents, languages, and medical histories not represented in training data can potentially hinder the performance of these systems, leading to poor patient outcomes.
4. Amazon Debuts Experimental AI Shopping Assistant
Amazon has introduced an experimental AI-powered shopping assistant that allows consumers to ask questions about specific items sold on the platform through the mobile app.
The “Looking for specific info” tab, previously displaying product reviews and answers to common questions, has been replaced with a large language model.
This AI-driven system summarizes information from the product listing page, enabling consumers to seek specific details about the item of interest.
However, the chatbot does not provide product comparisons, suggest alternatives, or perform actions such as adding items to shopping carts or disclosing pricing history.
An Amazon spokesperson confirmed the testing of the chatbot, highlighting its potential to improve shopping experiences by answering common product-related queries.
Similar to other chatbots, Amazon’s system is prone to occasional hallucination, and its responses should be taken with caution.
Intriguingly, the virtual shopping assistant exhibits diverse capabilities, including writing jokes, poems, and even generating code based on product information across multiple languages.
This article provides a comprehensive overview of recent advancements and challenges in the field of AI. For more in-depth information, explore the provided references and reputable sources.