Directly connected to Colab, supporting 20 programming languages: Is the Google version of ChatGPT code level anti-kill?

Although it has been leading the world's AI progress for more than ten years, Google has only been able to catch up with Microsoft and OpenAI in the past few months. For this reason, it even merged Google Brain and DeepMind this week. .

At the end of last year, ChatGPT triggered drastic changes in the technology industry. In February this year, Google released Bard, a ChatGPT competitor, and people received mixed reviews for its experience. Bard has limitations of one kind or another, and what many people ask of developers is "When can it write code?"

This Friday, Bard's ability to write code is finally online.

Google said that Bard can now use 20 programming languages, including C, Go, Java, Javascript, Python and Typescript, etc., and can also Help users debug, explain and optimize code. With just a simple prompt, such as "Can this code be made faster?", Bard can understand human intentions and automatically make the code more efficient.

We know that many people use Google Colab to run machine learning models, and it also comes with free cloud GPU computing power. Now people can also easily export Bard-generated Python code to Google Colab — without even copying and pasting. Bard can also help write functions for Google Sheets.

Previously, Google had announced that Bard would be open to users in the United States and the United Kingdom. These users can already directly use all the new features of Bard.

Google demonstrates the effectiveness of Bard writing code. Like ChatGPT, Bard can now generate code to complete the corresponding tasks according to your needs:

The function of explaining the code is special for programming beginners Useful:

In addition to generating and interpreting code, Bard can also help users debug code, including code generated by Bard itself. If the code generated by Bard doesn’t work as expected, just tell Bard: “This code didn’t work, please fix it (this code didn’t work, please fix it)” and Bard can help the user debug it.

Applying generative AI to accelerate software development and help people solve complex engineering challenges is Bard’s vision for launching code generation capabilities. But Bard's abilities still need to be improved.

Google says that Bard is still in its early experimental stages and may sometimes provide inaccurate, misleading, or false information. It may also generate code that does not produce the expected output, or that generates code that is not optimal. Excellent/incomplete code. Users need to carefully check the code, test and review the errors and bugs in the code before adopting the code generated by Bard.

For a large language model that is being tested on a large scale, newly launched functions are bound to encounter all kinds of teasing from users, and the same goes for Bard’s coding capabilities.

First of all, Bard will provide referenced code links when giving answers. For a practical product, this is very important and has been well received.

#You often complain that AI is copying code, it is not copying randomly.

Ask Bard "Can you help me implement a basic RNN and test it on dummy text data?" Then export the generated code directly to Google Colab. Part of the code doesn't work. Finding the error and asking the AI ​​again, Bard modified the code and everything seemed to be working perfectly. Now we just need to check that the implementation is correct, manually check it, and do some unit testing if necessary.

Having the Google Colab export feature is really useful.

Finally, someone tried to use Bard to generate code for the ancient programming language COBOL, and the result was actually satisfactory:

People have been worried that when this wave of COBOL programmers retire, many key positions will be left unfilled. It seems that AI can help us solve this big problem.

However, some netizens said that Bard’s ability still seems to be inferior to GPT-4.

Can using AI-assisted programming finally change the way we work? This remains to be explored.

The above is the detailed content of Directly connected to Colab, supporting 20 programming languages: Is the Google version of ChatGPT code level anti-kill?. For more information, please follow other related articles on the PHP Chinese website!