JAKARTA - Google is again competing to improve sophistication in its Artificial Intelligence (AI) chatbot, Bard with better reasoning and math capabilities.
To achieve that goal, Bard will no longer rely solely on the Big Language Model (LLM) but on a technique called implicit code execution.
LLM like Bard is basically a prediction machine. When given the prompt, they generate a response by anticipating what words might appear next in a sentence.
Making the tool a very good email and essay writer, but a software developer who is prone to error.
With the execution of the implicit code, Bard identifies requests that may benefit from logical codes, write them down, execute them, and use the results for more accurate responses.
So Bard will now create and execute Python code in the background whenever users ask questions that require computation, such as calculating the prime factor of a number, or turning the character series around.
"So far, we have seen this method increase Bard's accuracy of the response to computing-based word and mathematics in our internal challenge dataset by about 30 percent," said Bard Product Leader Jack Krawczyk and Technical Vice President Bard Amarnag Subramanya, in a company blog post, quoted Thursday, June 8.
また読む:
According to both, even with this upgrade, Bard doesn't always do it right. For example, chatbots may not produce codes to help with a quick response, the code they generate may be wrong, or Bard may not include a code executed in its response.
"With all that, increasing the ability to respond with structured and logic-driven capabilities is an important step to make Bard more useful," said Krawczyk and Subramanya.
In addition, when Bard responded to the command with a table, users can now export the table directly to Spreadsheet.
Transferring this data to Spreadsheet can help users start editing it immediately, instead of manually copying it and risk losing formulation or data in the process.
The English, Chinese, Japanese, Arabic, and French versions are automatically generated by the AI. So there may still be inaccuracies in translating, please always see Indonesian as our main language. (system supported by DigitalSiber.id)
