Organize | Kexin Zhu
Produced | CSDN Program Life (ID: coder_life)
At the beginning of the year, the major update of GitHub Copilot made countless programmers call out: "The era of GPT-4 programming is here!"
Today, GitHub Copilot has been upgraded again.
Continuous improvement on the old model: coding speed, efficiency improvement
The efficiency of coding has always attracted the attention of programmers.
Earlier, Github CEO Thomas Dohmke put down his words: "Sooner or later, 80% of the code will be written by Copilot!"
In October last year, Thomas Dohmke made relevant predictions.
It is estimated that within the next five years, Github Copilot will generate 80% of the code . He then also claimed that 40% of the code had been generated in the Github Copilot beta, resulting in a 55% speedup for developers.
Today, the advanced upgrade of GitHub Copilot has indeed lived up to expectations.
At the beginning of the year, GitHub Copilot switched from the GPT-3-based "Codex" to the more advanced AI model GPT-4. Under the change, the reasoning and code generation capabilities of the new model have been significantly improved.
Today, the new AI model has been upgraded again, reducing latency by 13% compared to the previous Codex model .
In simple terms, this means that GitHub Copilot can generate code suggestions faster than ever before, increasing developer productivity.
It is reported that the new AI model was jointly developed by OpenAI, Azure AI and GitHub.
Enhanced contextual filtering so developers can focus on being creative
In addition to the improved AI model, GitHub Copilot has enhanced contextual filtering.
Helps developers receive suggestions that are more relevant to their specific coding tasks by filtering hints and code suggestions more intelligently, resulting in a 6% relative increase in code acceptance rates .
In other words, during the coding process, developers can focus on the creativity of their work instead of getting bogged down in tedious coding tasks.
According to GitHub, this helps developers work more efficiently without changing their workflow, which ultimately leads to innovative, high-quality code.
It is worth noting that the "new upgrade" has a wider scope of application, and both personal and enterprise versions can be used.
At present, other updates for GitHub Copilot, GitHub officials said, will be announced at the upcoming GitHub Universe event.
Github CEO: "This doesn't mean developers will be replaced!"
Needless to say, improvements to GitHub Copilot benefit developers by minimizing disruption and enabling faster innovation.
But when the traditional coding habits are changed, there will undoubtedly be an anxiety of "worried about being replaced" .
For this reason, many developers lamented:
"Copilot can speed up writing code and reduce effort, worrying about being replaced";
"Really good programming development companion, Copilot makes me anxious by its ability";
"As a developer's assistant, I hope it is powerful but I am afraid that it is too powerful"...
In June of this year, in an interview with Freethink, Thomas Dohmke also tried to "answer questions" for everyone.
In addition to explaining how AI will change the way we code, work and learn, Thomas Dohmke also conveyed his views on how Copilot affects developer productivity.
Thomas Dohmke believes that even if 80% of the code will be written by Copilot, this does not mean that developers will be replaced, but developers have more time to focus on the 20% they are writing .
Additionally, in a quantitative survey of 2,000 developers, Thomas Dohmke found that 75% of developers said they felt more fulfilled using Copilot.
At the same time, they also conducted a qualitative study of 100 people, divided into two groups - 50 developers who were given access to Copilot, and the remaining 50 who did not have access to Copilot, and each group was asked to build a web server .
In the end, the group equipped with Copilot was 55 percent faster and had a 78 percent success rate , higher than the group without access to Copilot.
Reference link:
https://web.archive.org/web/20230728192619/https://github.blog/2023-07-28-smarter-more-efficient-coding-github-copilot-goes-beyond-codex-with-improved-ai-model/
https://the-decoder.com/new-ai-model-boosts-github-copilots-code-generation-capabilities/
https://ts2.space/en/github-copilot-gets-upgraded-ai-model-and-enhanced-contextual-filtering/
https://www.freethink.com/robots-ai/github-copilot
