Google Improves Bard: AI Chatbot for Logic and Code
Bard, Google’s chatbot, continues to evolve and now handles tasks related to logic and reasoning. In its latest blog post, the tech giant announced significant progress in this area. Thanks to a new technique known as “stealth code execution”, Bard has greatly improved his math and coding skills.
Chatbots based on large language models (LLMs) like Bard use a “prediction engine” method. They generate clue-based responses by predicting which words are most likely to follow in a sentence. While useful for writing emails and essays, it is ineffective for writing code.
To overcome these limitations of large language models, Google has developed a “hidden code execution” technique that allows Bard to write and execute its own code. In the latest version of Bard, it is able to identify useful hints, write code under the hood, test it, and use the results to generate more accurate answers.
According to internal testing, the new Bard improved its responses to verbal and math problems by 30% compared to the previous version of the chatbot. However, the developers have noted that even with these improvements, Bard will not always run flawlessly. In some cases, Bard may not generate code, generate code incorrectly, or not include executed code in the response, notes NIXSolutions.
When Bard launched earlier this year, it didn’t look much different from other chatbots like Bing Chat and ChatGPT. Google employees who tested Bard prior to its release expressed serious concerns about its reliability, calling it a “pathological liar” and “worse than worthless.” However, Google is looking to change that with updated code generation and other improvements to Bard, including support for new languages, multi-modal queries, and image generation.