Google has given its chatbot Bard an update that allows it to generate code in over 20 programming languages. The update also brings a feature that sets Bard apart from ChatGPT.
In addition to writers who string sentences together in natural language, one audience, in particular, has benefited from the progress of large language models: programmers. There are many reports of users who have been able to implement their own software ideas using ChatGPT or other language models without any real programming knowledge.
Often a basic understanding of the syntax is helpful, and it is not uncommon for the AI to take several tries before it produces error-free code. Nevertheless, language models are a very helpful tool.
More than 20 programming languages, including Google Sheets syntax
Code written in Python, which plays a dominant role in machine learning, can be exported with a click to a Google Colab notebook (similar to a virtual machine for running Python in the browser). Conveniently, Google’s AI also masters Google’s own syntax for the spreadsheet program Sheets.
In addition to generating code, Bard can also explain a program line by line, debug it, or simply improve it, for example, to reduce runtime.
Bard shows open-source project sources
Paige Bailey, a product manager at Google Research, stresses that Bard’s coding capabilities are still in the early stages of development, so the AI may spit out “inaccurate, misleading or false information while presenting it confidently.”
In this respect, Bard doesn’t really differ from ChatGPT and the like. Unlike OpenAI’s previous solutions, however, Bard does provide a source when it takes large chunks of code from an open-source project, according to Bailey.
All in all, the new Bard features should not be seen as a “bonus”, but rather as an absolute necessity: A language model without code generation is unlikely to be successful in the long run. However, in some cases, the ability to check sources could make the work of programmers easier.
How Bard’s programming capabilities compare to other language models remains to be seen. The expansion of the feature set is not accompanied by an expansion of the user base, as Bard is currently in a closed beta. However, the recently improved logic and math capabilities should provide a good basis for usable code.