ChatGPT came into our lives at the end of last year and since then it has been non-stop in the segment of generative Artificial Intelligence models. Such has been the upheaval that even Google has had to make a move to keep up, despite the fact that the company has spent years investing in the development of Artificial Intelligence systems.
As we already mentioned at the time, Google Bard is Google’s response to the rise of ChatGPT and similar services based on OpenAI technologies. Google’s chatbot, currently considered an experiment and not a finished product, has received different improvements in these short months of life.
The improvements that are present today
Today, precisely, it receives two other improvements, according to Google advances in a statement.
The search giant says Bard is better as of today at math homework, coding questions, and string manipulation. And in addition, now it is also capable of exporting the generated tables directly to the Spreadsheets application.
Google gives as an example that if:
Bard generates a table in his response, just like if he asks you to “create a table for volunteer signups for my animal shelter”, now you can export it directly to Sheets. .
In this regard, it should be remembered that last May it acquired export capabilities to Gmail and Docs, so it was a matter of time before it continued integrating actions that make it more compatible with most Google services.
Making it possible to offer more precise and reasoned answers
When it comes to improvements to math tasks, coding questions, and string manipulation, Google says that this is now possible thanks to the arrival of a new technique called “implicit code execution,” which helps the Google chatbot in detecting computational hints and running code in the background, allowing you to deliver results with greater accuracy.
Google explains that long language models are quite good in areas like language and creative tasks, but not in areas like reasoning and math, so relying on long language models is not enough, so with The arrival of the new technique allows Google Bard to improve his math and reasoning skills.
The way in which it has been possible to improve your logical skills
And building on the dichotomy of human intelligence laid out in Daniel Kahneman’s book “Thinking, Fast and Slow”: the separation of “System 1” and “System 2” thinking, Google says that extended language models work in the System 1, generating texts quickly without delving into them, while traditional calculus closely aligns with System 2 thinking, which he describes as formulaic and inflexible, which when used correctly is capable of delivering impressive results.
And under this mode of work, they point out that:
we have combined the capabilities of LLM (System 1) and traditional coding (System 2) to help improve the accuracy of Bard’s responses. Through implicit code execution, Bard identifies prompts that could benefit from logical code, writes them “under the hood,” executes them, and uses the result to generate a more accurate response.
As a result, Bard has been able to improve by up to 30% in accurately answering word and math problems based on calculations from a data set, although Google notes that Bard will not do well in all cases. , for example, when generating code to help provide a quick response, although in any case it makes Bard more useful.
Logically, from Google they already leave the doors open to future improvements in the precision of the responses, so we will have to wait to see the official announcements about the methods used to progress in this regard.
More information: Google
