Google's Bard AI empowered to generate and execute code for question responses

Google's recent strategies target to enhance the computational capabilities of language models like ChatGPT and Google Bard. These models now write and execute programs to deliver accurate answers rather than providing flawed results for math or string manipulation tasks. 

When presented with a computational question, in place of displaying the output directly, the language model generates a program and gives the program's output as the answer. 

For example, when asked to reverse the word "Lollipop," ChatGPT falls short and responds with the incorrect answer "pillopoL" because of its restricted comprehension of word chunks or tokens. In contrast, Bard correctly reverses the word to "popilloL" and even includes the Python code it generated to arrive at the solution. While programming enthusiasts may find the behind-the-scenes code intriguing, regular users who strive for straightforward answers to their queries may find it confusing and irrelevant.

Google establishes a comparison between AI models writing programs and humans performing long division, highlighting the distinction between fast, intuitive "System 1" thinking and slow, deliberate "System 2" thinking. 

Language models primarily operate under System 1 thinking, which delivers extraordinarily but sometimes results in poor outcomes. On the other hand, traditional computation aligns with System 2 thinking, offering formulaic yet reliable results through a series of well-defined steps, similar to solving long division problems, according to Daily Magazine.