Listen to this story
From sci-fi dreams to real-life implications, the evolution of AI has become a global sensation. Its potential to boost the global GDP by 7% is huge, and yet its impact on over 300 million jobs in major economies is equally threatening. But amid the fear of repercussions, developers remain unfazed.
Recently, Chris Albon, director of ML at Wikimedia (the nonprofit host of Wikipedia) took to Twitter to share that ChatGPT makes his coding faster. However, he admits that ChatGPT is bad at “big things” like architecture or the product’s concept or goals but can give a “a working/near-working solution in one second”.
Satya Mallick, the chief executive officer at OpenCV, an open-source computer vision and machine learning software library, told AIM, “We should look at AI as a collaborator instead of a competitor that humans can leverage to be more productive”
ChatGPT Over Copilot?
Autocoder GitHub Copilot has been a fan favourite since its release last year. But with a chat-based interface coming into picture, developers are shifting to ChatGPT. Although Copilot can offer suggestions for lines of code, variables, and function names and can be integrated with the IDE and assists with small code snippets, Copilot has its limitations, such as producing inaccurate or inefficient code proposals and being inadequate for complicated programming duties that demand extensive knowledge and expertise.
On the other hand, ChatGPT is easier for generating code answers and clarifying code ideas as it has a conversational interface. Not just newbies, ChatGPT increases efficiency of tech veterans too, as echoed by iOS engineer Nathan Chan. Additionally, it can aid in the creation of function and unit tests to improve test-driven development techniques. But it is not fitting for time-sensitive applications such as gaming or trading.
Nonetheless, GitHub is set to upgrade its Copilot by introducing CopilotX that is built on GPT-4 just like ChatGPT and will provide an experience similar to ChatGPT right within the VS Code IDE. In addition to suggesting code, CopilotX offers comprehensive code analysis, suggests bug fixes, and identifies the code entered by the programmer. Furthermore, GitHub is developing a voice-to-code model capable of transforming spoken language instructions into code.
Read more: GPT-4 Comes To GitHub With Copilot X
LLMs Can Now Self-Debug
Besides generating codes, LLMs can now also debug on their own with ease, which has been considered to be a long-standing obstacle.
According to a recent Google Brain publication, LLMs not only generate codes but also possess the ability to self-debug their codes. This has resulted in a nearly 10% improvement in accuracy. LLMs’ natural language processing capability allows them to identify and rectify mistakes in their work. By implementing this method, LLMs have achieved optimal performance across various code generation evaluations, producing outputs of greater value compared to those without self-debugging.
Debugging is a major obstacle in the workflow, and improving self-debugging capabilities is crucial for enhancing LLM coding. Integrating autocoding platforms with other AI approaches, such as constitutional AI, Reflexion, and AutoGPT, can significantly enhance productivity.
ChatGPT is Killing Stack Overflow
Prior to the arrival of ChatGPT, programmers depended on Stack Overflow to seek solutions from the online developer community. However, in December of 2022, ChatGPT was barred from Stack Overflow citing falsity over the AI-generated responses, resulting in a 12% reduction in the total number of website visits in December. The number of visits decreased from 279 million in November 2022 to 247 million by the end of December.
But when AIM reached out to Stack Overflow to understand about the drop, they dismissed the impact of ChatGPT on traffic. “Like many other technology sites during the holidays, the month of December often brings a seasonal shift as users (many of whom are professional developers) take time off to be with their friends and families. Our public platform continues to serve 100 million visitors every month, making it one of the most popular websites in the world,” Stack Overflow told AIM.