Google Bard Better At Logic And Reasoning, With Improved Math, Coding And String Manipulation

Google Bard

Logic has always been the thing behind computers. The thing is, computers can have a hard time explaining its logic.

Thanks to AI, computers have become smarter that they can be tasked to do something beyond their programming. And thanks to OpenAI's introduction to ChatGPT, the generative AI is the hype the AI industry had longed for.

And thanks to generative AIs, humans can finally "ask" AIs to explain their logic, in a way humans would.

Google has what it calls Bard, and that it's a direct competitor to ChatGPT.

Unlike the free version of ChatGPT, Bard is connected to the internet, thus making it capable of answering questions its rival cannot.

After giving it the ability to debug code, this time, Google is improving Bard significantly, by improving Bard's "mathematical tasks, coding questions and string manipulation through a new technique called implicit code execution."

In a blog post, Google announced two updates to Bard.

The first update, is "better responses for advanced reasoning and math prompts."

According to Google, Bard has been updated with a technique called "implicit code execution," which should help the generative AI detect computational prompts and run code in the background.

As a result of this, Bard can respond more accurately to mathematical tasks, coding questions and string manipulation prompts.

Google provides the example input of "Reverse the word 'Lollipop' for me."

For ChatGPT, this be an issue, because the answer it provides is the incorrect answer "pillopoL." This can happen because generative AI' language models see the world in chunks of words, or "tokens." Because of this, they're not good at this of task.

But Bard answers it with the correct "popilloL."

This is done because Google likens the AI model to writing a program to humans,

This implicit code execution technique enables Bard to detect computational prompts and run code in the background, allowing it to answer questions like finding prime factors, calculating growth rates, or reversing words more effectively.

Large language models like Bard excel at language-related tasks, but they often struggle with more complex problems that require reasoning and math skills.

The goal behind implicit code execution technique is to equip Bard with advanced reasoning and logic capabilities.

Previously, Bard operated primarily under "System 1" thinking, producing responses quickly but without deep thought. By incorporating traditional code execution ("System 2") into its process, Bard is able to enhance the accuracy of its responses.

According to Google, its method has shown promising results, improving the accuracy of Bard's responses to computation-based problems by approximately 30% in internal tests.

What's more, Bard can also include a python code it used to answer the question. This is a way for Bard to seemingly open its hood, and show how its underbelly works.

The second update that follows this, is the ability for Bard to export users' actions to Google Sheets.

"So when Bard generates a table in its response — like if you ask it to 'create a table for volunteer sign-ups for my animal shelter' — you can now export it right to Sheets," Google's blog post said.

Read: Google Bard And OpenAI ChatGPT 'Are Large Language Models, Not Knowledge Models'

Published: 
07/06/2023