Bard enhances logic and reasoning skills with implicit code execution

Google recently reported that Google Gemini is taking a significant leap forward in its logic and reasoning abilities through the implementation of an advanced technique known as implicit code execution. This method enables Bard to identify computational prompts and seamlessly execute code in the background, resulting in remarkably accurate responses to mathematical tasks, coding questions, and string manipulation prompts. For example, we should now expect better results to questions such as: 'What are the prime factors of 15683615?', 'Calculate the growth rate of my savings', or 'Reverse the word “Lollipop” for me'.

The improved accuracy of Bard's responses stems from combining large language models (LLMs) with traditional code using implicit code execution. Google drew inspiration from Daniel Kahneman's book "Thinking, Fast and Slow," which explores the distinction between fast, intuitive "System 1" thinking and slow, deliberate "System 2" thinking in human intelligence. System 1 thinking is effortless and instinctive, while System 2 thinking requires effort and deliberation. Bard's fusion of LLMs and traditional code harnesses the strengths of both thinking systems to provide more accurate responses. LLMs operate under the "System 1" thinking process, characterized by its speed and intuition, while traditional code aligns with the more deliberate and slower "System 2" thinking. By harnessing the strengths of both systems, Bard can generate and execute code, enabling it to provide precise and reliable answers.

Initial internal tests using challenging datasets have demonstrated a remarkable 30% improvement in the accuracy of Bard's responses to computation-based word problems and math queries. However, it is important to note that while Bard has shown exceptional proficiency in generating and executing code, there are instances where it may not generate code or include the executed code in its response. It's crucial to exercise caution, as there remains a possibility of incorrect code generation.

Another notable enhancement introduced in Bard's latest update is the integration of a new export action, allowing users to directly export generated tables to Google Sheets. This feature streamlines the workflow for individuals who rely on Bard's capabilities to generate data and insights. With just a simple command, users can now export the generated tables to Google Sheets, saving time and effort in data management.

While ChatGPT and Microsoft Bing Chat currently holds the lead in AI-powered search assistance, Google's Bard is steadily closing the gap by expanding its functionalities and features. These recent advancements mark a significant milestone in enhancing Bard's capabilities for structured, logic-driven tasks. Through implicit code execution, Bard's logic and reasoning skills receive a substantial boost, enabling it to provide increasingly reliable and accurate responses to complex computational challenges.