A new technology called Codex generates programs in 12 coded languages and can even translate between them.
But it's not a threat to experienced programmers.
As soon as Tom Smith got his hands on Codex , a new artificial intelligence technology that could write its own computer programs — he gave it a job.
He asked if it could solve the "coding challenges" programmers often face when interviewing for high-paying jobs at Silicon Valley companies like Google and Facebook. Is it possible to write a program that replaces all spaces in sentences with dashes? Can I write a code that identifies invalid zip codes?
It does both at once, before completing several other tasks.
"These problems are difficult for a lot of people to solve, including myself, and it responds in two seconds ." Mr. Smith, a seasoned programmer who oversees an artificial intelligence startup called Gado Images Say. "It looks so scary."
Codex seems to be a technology that will soon replace human programmers. As Mr. Smith continued to test the system, he realized its skills went far beyond a knack for answering common interview questions. It can even be translated from one programming language to another.
Illinois Institute of Technology, a famous American school , will recruit students in the spring of 2022. The exclusive experience camp in China is here. Let’s experience the benefits of an online master’s degree in computer science! Two weeks for a limited time, limited to 30 people! The whole process is free , click the link below to lock the experience camp qualification!
Online Computer Master’s Experience Camp
However, after using the new technology for a few weeks, Mr Smith doesn't think it poses a threat to professional technicians. In fact, like many other experts, he sees it as a tool that will ultimately increase human productivity . It could even help a new generation learn computer art by showing them how to write simple snippets of code, like a personal tutor.
"It's a tool that makes a programmer's life easier," Mr. Smith said.
▲ Testing Codex convinced Mr. Smith, who runs an artificial intelligence start-up, that it would only increase the efficiency of people's use of computers. Jason Henry for The New York Times
Codex is built by OpenAI , one of the most ambitious research labs in the world, providing insight into the state of artificial intelligence.
Despite the leaps and bounds of broad AI technology over the past decade, even the most impressive systems end up being human tools, not replacing them .
Thanks to the rapid rise of mathematical systems called neural networks, machines can now learn certain skills by analyzing large amounts of data. For example, they can learn to identify cats by analyzing thousands of cat photos.
About four years ago, researchers at labs like OpenAI began designing neural networks for analyzing vast amounts of prose, including thousands of digital books, Wikipedia articles and various other texts posted to the internet.
By identifying patterns in all text, the network learns to predict the next word in the sequence. When someone enters a few words into these "universal language models," they can complete the idea with an entire paragraph. In this way, an OpenAI system called GPT-3 can write its own Twitter posts, speeches, poems and news articles.
To the surprise of the researchers who built the system, it can even write its own computer programs, albeit short and simple . Apparently, it learned from countless programs posted on the Internet. So OpenAI went a step further and trained a new system — Codex — with lots of prose and code.
, duration 00:18
▲ If you ask Codex to "make a blizzard on a black background", it will do so, generating and running the code.
The result is a system that understands both prose and code — to a degree. You can type in plain English, like snow on a black background, and it will give you the code to create a virtual snowstorm. If you ask for a blue bouncing ball, it will give you that too.
"You can tell it to do something and it will do it," says Ania Kubow, another programmer using the technology.
Codex can generate programs in 12 computer languages and can even translate between them . But it often makes mistakes, and while its skills are impressive, it's a pity that it can't reason like a human. It can recognize or imitate what it has seen in the past, but is not flexible enough to think for itself.
Sometimes programs generated by Codex fail to run. Or they contain security holes. Or, they are far from what you want them to do. OpenAI estimates that Codex has a 37% chance of generating the correct code.
In other words, Codex is only really useful for experienced programmers .
But it can help programmers do their day jobs faster. It can help them find the basic building blocks they need or point them to new ideas. Using this technology, GitHub acts like the way Autocomplete tools suggest the next word as you type text or email.
"It's a way of writing code without writing as much code as possible," said Jeremy Howard, founder of the artificial intelligence lab Fast.ai, who helped create the language OpenAI's work is based on Technology. "It's not always correct, but the accuracy is acceptable."
, duration 00:12
▲ Codex created a website for "A Cat Is a Lawyer" with a biography, phone number and a small profile picture.
Mr. Howard and others believe Codex can also help newbies learn to code . It is particularly good at generating simple programs from short English descriptions. It also works in the other direction by explaining complex code in plain English. Some, including Swedish entrepreneur Joel Hellermark, are already trying to turn the system into a teaching tool.
The rest of the AI field looks similar. Robots are getting more powerful. The same goes for chatbots designed for online conversations.
DeepMind, a London-based artificial intelligence lab, recently built a system that instantly recognizes the shape of proteins in the human body, a key part of designing new drugs and vaccines. This task used to take scientists days or even years.
But these systems replace only a fraction of what human experts can do.
There are only a few areas where new machines can immediately replace workers, such as shipping centers.
▲ OpenAI's Greg Brockman said that artificial intelligence is eliminating the drudgery of work, not replacing it. Steve Jennings/Getty Images
With the launch of his startup, Gado Images, Mr. Smith set out to build a system that could automatically organize newspaper and library photo archives, redisplay forgotten images, automatically write titles and tags, and collaborate with other publications and Businesses share photos. But the technology can only handle part of the job.
It can sift through vast photo archives faster than humans, identifying potentially useful types of images and inserting captions. But finding the best and most important photos and tagging them properly still requires an experienced archivist .
"We had thought that these tools would completely eliminate the need for humans, but years later we learned that this is actually not possible - you still need a skilled person to review the output," Mr Smith said. "Technology can be biased. You still need a human to review what it does and decide what's good and what's not."
Codex expands the capabilities of the machine, but it shows again that the technology works best under human control.
Greg Brockman, CTO of OpenAI, said: "AI is not working the way anyone expected it to be."
"It's not replacing anyone's job, it's taking everyone's drudgery away."