Friend or foe: Can computer coders trust ChatGPT?

    Pietro Schirano asked ChatGPT to create Pong, Asteroids and Breakout for him

 "It felt like magic to me," says Pietro Schirano. "It scared me a little bit because it was so good."

He is referring to the most recent version of the artificial intelligence (AI) tool ChatGPT. You type something in, and it makes up an answer for you.

The responses appear to be friendly, intelligent, and human, which is strange. GPT-4 is an artificial intelligence model beneath it.

Mr. Schirano, the design lead at the business finance company Brex, has been developing computer programs with ChatGPT.

He wrote, referring to the 1970s table tennis game from Atari, "I want to build a game like Pong." For a large number of people to try, what is the best language to use?"

ChatGPT not only suggested JavaScript, a browser-based programming language, but it also generated the code for a simplified version of the game.

ChatGPT wrote in 40 seconds what would have taken a skilled developer half an hour to write. Breakout and Asteroids, two classic arcade games, performed similarly to it.

    Pong, seen here being played at a trade fair in Cologne in 2019. ChatGPT quickly created a simple    version of the game

Although there are numerous online code examples for these games, ChatGPT is not searching the internet for a pre-made solution.

ChatGPT generates new text specifically for you by predicting the correct answer, despite the fact that it was constructed using a large amount of web content.

According to Mr. Schirano, "ChatGPT was able to understand my question about how I make this game more popular." It has a lot of contextual understanding that a person who has never developed would not be able to get from Google.

You would need to use a search engine to determine which coding language to use, how to use it, and how to complete any online examples that are incomplete. ChatGPT can give you all the game code, along with directions custom-made to your requirements.

Mr. Schirano asserts, "Now, even someone who has no idea what they're doing can create a simple web app."

Because there are so many examples online, some claim that replicating these older games was too simple.

Therefore, Brex design manager Ammaar Reshi, a colleague of Mr. Schirano, advanced ChatGPT. He requested that it recreate a previous PC game for which there was no online example code. Skyroads is a 3D game in which you steer a car along a road while jumping over obstacles.

The game's Wikipedia page informed ChatGPT, and Mr. Reshi provided additional information regarding the game's operation. "With this one, there was so much back and forth," says Mr. Reshi. "About an hour of tweaking this and that." He had to play the game, demonstrate how to alter the game's behavior to ChatGPT, and repeat.

Despite this, Mr. Reshi was able to create an online 3D game despite having no prior knowledge of the JavaScript programming language or having access to code examples for that game.

These are little games, and no one's proposing ChatGPT will make current control center games. However, it could be put to use in the creation of major game components or any other software.

However, even with the efficiency enhancements, some wonder whether or not to embrace artificial intelligence for coding.

Tony Smith, chief technology officer of Rightly, a company that helps individuals get their data deleted from company databases, asserts that "the risks of ChatGPT for software development still outweigh the benefits." The code that these AIs produce frequently contains errors or is inefficient.

Mr. Smith requested that ChatGPT develop some code to determine the number of days in a given month. "It thinks March has 30 days because there is a problem with the switch to British Summer Time," he says. "The code looks great." Once per year, a subtle bug will surface."

Others have discovered that security flaws can be present in AI-generated code. ChatGPT may use out-of-date methods, overlook vulnerabilities, or model bad behavior it sees online. Additionally, ChatGPT has limited information regarding future events.

ChatGPT has been blocked by the Italian data protection authority while it looks into whether it followed the General Data Protection Regulation.

    Kevin Bocek says developers must be accountable for AI-generated code

People who are tempted to use code they don't understand also increase business risk.

Venafi, a company that develops security software for authenticating machines, employs Kevin Bocek as vice president, security strategy and threat intelligence. He used ChatGPT to create PowerShell scripts and Excel macros, two different ways to give a computer instructions that can be repeated. They are often used by hackers to launch attacks.

He asserts, "Now I've got code that's come from "out there," and I'm going to run it, but I've got no idea what it does." The potential for additional potentially harmful code to enter has now increased."

Even though AI can be used to create code, it will always be the developer's job to check it at Venafi. He states, "That's something we take very seriously." Humans review code multiple times. At last, the expert engineers are dependable and responsible."

    Matthew Hunt (second from left) asks whether ChatGPT might stop him growing his skills

On some of its websites, the performance marketing and web agency Herd creates slideshow-like elements. Composing the JavaScript code independently for each part can be difficult, so ChatGPT truly speeds that cycle up," says Matthew Chase, a lesser web designer at Crowd.

However, he is concerned that ChatGPT might also eliminate some of the satisfying work. Part of the delight of creating is realizing precisely exact thing a document does and how everything functions," he says. " We aren't really developing as developers if ChatGPT is doing this for us."

D2iQ vice president of technology Dan Ciruli has a different perspective. He used to work at Google.

Developers can connect their software to Google services by using the application programming interfaces (APIs) that Google provides. There were 200 APIs and seven upheld programming dialects, including Python. Google would need to create 1,400 code examples in order to provide instructions for each API in each language.

He states, "That's something we struggled with." We simply were unable to produce samples in every programming language. Yet, when I gained admittance to ChatGPT, the principal thing I did was request that it tell me the best way to call the Google Decipher Programming interface in Python. It was there, boom."

One well-known AI tool's name suggests that the future of software development will likely involve a partnership between the developer and the AI assistant. A commercial tool called GitHub Copilot automatically completes code as developers work. A model that is quicker but less accurate than ChatGPT is used for the code completion.

According to Thomas Dohmke, CEO of GitHub, "Given you're typing code, ideally every time you input a character, you get a new response."

"Humans continue to be an essential safeguard, even though Copilot has been upgraded with features that can detect and block insecure code." Like some other coding apparatus, you ought to continuously utilize GitHub Copilot along with human code survey, testing practices and security instruments," says Mr Dohmke.

Will the creation of AI code result in job losses? Mr Schirano and Mr Reshi can both predict that lesser jobs could be supplanted assuming man-made intelligence takes on tedious coding work.

Manuel Doc works as a front-end developer at Illustrate Digital, a user experience company. I was so stunned by ChatGPT's code-level reactions that I was terrified my work would be in peril soon," he says.

"However, during my one-on-one meeting with our head developer, he helped me realize that I'm not just hired to write code; I'm also hired to analyze issues and offer solutions that ChatGPT cannot. It was extremely comforting."