Google’s mysterious project exposed! An AI that can write code and fix bugs makes coders tremble

How stressful is it to be a programmer?

Anyway, a recent rumor is making coders tremble.

It is said that Google is developing a secret new project to teach AI to write code.

It is said that after learning, AI can not only write code, but also fix bugs.

When it comes to writing code, you should let the code do it by itself

It is rumored that this secret project of Google will use machine learning to train the code and let them code themselves. , fix bugs by yourself, and update by yourself.

This project shows that Google is one step closer to generative artificial intelligence.

Today’s AI is becoming more and more omnipotent. They can create images, videos, and write code.

If this AI continues to evolve, will coders still be needed in the future?

According to people familiar with the matter, this project was originally developed by Alphabet’s lunar landing unit, the X department, code-named Pitchfork.

This summer, it was moved to Google Labs.

As we all know, Google Labs values ​​“long-term investment”, including VR and AR projects.

Now, Pitchfork has become an employee of the "AI Developer Assistance Team" under Google Labs.

According to internal information, Pitchfork’s role is to “teach code to write and rewrite itself.”

It can learn different programming styles and write code according to these styles.

Now, the team is exploring different use cases to help developers.

A Google employee said that the original intention of developing Pitchfork was to build a tool to update Google's Python code base to a new version.

How do you transition from one version to the next without hiring extra software engineers?

Pitchfork came into being.

Team leader Hatalsky said that over time, the goal of the Pitchfork project gradually became to build a universal system.

Since the end of last year, Pitchfor has been able to reduce the cost of for X.

AlphaCode: Beating 72% of programmers

In fact, AI programming is not new for a long time.

In February 2022, DeepMind, another subsidiary of Alphabet and a sister company of Google, launched a system called “AlphaCode” that can use artificial intelligence to generate code.

According to DeepMind, AlphaCode can rival humans.

DeepMind tested AlphaCode using 10 existing competitions hosted on the programming competition platform Codeforces, and ranked in the top 54.3% overall, meaning it beat 46% of the entrants.

DeepMind claimed that AlphaCode solved 34.2% of the problems in 1 million samples when tested using the programming competition platform Codeforces.

In addition, among the users who have participated in the competition in the past 6 months, AlphaCode’s data ranks in the top 28%. It can be said that "it beats 72% of human programmers" ”!

At that time, DeepMind pointed out that although AlphaCode is currently only suitable for competitive programming fields, it is obvious that its future capabilities will not stop there.

It opens the door to the creation of tools that will make programming more accessible and one day fully automated.

Copilot: A good helper for programmers

Moving forward, in 2021, GitHub and OpenAI jointly launched an AI programming artifact-GitHub Copilot.

When entering code, Copilot will automatically prompt the code snippets that may appear next in the program, just like a person who has been trained to speak in Python or JavaScript. Autocomplete bot.

Copilot can fill in necessary code blocks, as long as they are not particularly complex or creative, which is very useful for programming that amounts to manual labor.

In addition, Copilot also optimized the online collaboration function between multiple programmers, so it was one of the most successful projects in the early days of generative AI.

On June 22, 2022, Copilot was officially launched for the C side, priced at US$10/month or US$100/year, and is provided free of charge to student users and maintainers of popular open source projects .

Now, thousands of developers are using Copilot.

Up to 40% of code written in a dozen of the most popular languages ​​relies on it to generate.

GitHub predicts that developers will use Copilot to write up to 80% of their code within five years.

Microsoft Chief Technology Officer Kevin Scott also said: "We are convinced that GitHub Copilot can be applied to thousands of different types of work."

However, due to alleged infringement, Copilot was sued by angry programmers in court less than five months after its release, claiming US$9 billion.

Codex: AI real-time automatic programming

OpenAI pioneered text generation.

Starting in 2019, OpenAI began to use an algorithm called GPT-2, which caused a sensation in the industry; at the end of 2021, OpenAI launched an upgraded version of GPT-2, GPT-3. Available to anyone.

GPT-3 has 175 billion parameters, which is 100 times that of the previous generation model GPT-2. It also improves the previous parameter record of similar NLP models by 10 times.

In the field of image generation, OpenAI officially announced DALL-E in January 2021, which can generate original images for text prompts.

In April 2022, DALL-E 2 was released, which is capable of rendering more complex images.

On June 30, 2021, OpenAI and GitHub jointly released the "AI code completion artifact" GitHub Copilot by carrying their own Codex model .

However, at that time, Codex did not reveal too many details and always maintained a sense of mystery.

On August 10 last year, OpenAI finally launched an improved version of Codex and also released a private beta version based on its own API.

Compared with the previous version, the improved version of Codex can not only interpret simple natural language commands, but also automatically create and complete codes, making it more flexible and advanced.

For example, in OpenAI's space game "space game", the user inputs the natural language command "Make it be smallish", and the Codex system will automatically generate the control code to make the spacecraft in the picture The size is reduced.

In addition, there is a magical tool that can write soft articles by yourself - Jasper.

"Jasper" is an AI content platform equipped with GPT-3's text generation technology, which can help humans break through creative barriers and automatically generate text for company use at 10 times the speed. , for marketing, blogging, email, and more.

Just in October, Jasper announced that it had raised US$125 million in funding, with a current valuation of US$1.5 billion, and claimed that it was expected to bring in US$75 million in revenue this year.

From last year to this year, generative AI has exploded

The hottest word in the industry right now is "generative artificial intelligence" "Smart" is definitely one of them.

In technical terms, “generative artificial intelligence” refers to unsupervised and semi-supervised machine learning algorithms that enable computers to use text, audio and video files, images, and even code to create new content.

The artwork, text and code generated by AI based on user prompts have amazed humans time and time again.

#Gartner listed generative AI as a technology that can bring about a productivity revolution in the "2022 Emerging Technologies and Trends Impact Radar Report" one.

According to Gartner predictions, generative AI will generate 10% of all data by 2025 (now less than 1%), as well as for 20% of all test data for consumer use cases.

And, by 2025, 50% of drug discovery and development will use generative AI.

Large biopharmaceutical companies investing in artificial intelligence

And by 2027 In 2019, 30% of manufacturers will use generative AI to improve product development efficiency.

Now, generative AI has triggered a “gold rush” in Silicon Valley.

In a recent blog post, the venture capital firm Sequoia Capital elaborated on the potential of generative artificial intelligence in fields such as speech synthesis, video editing, biology, and chemistry.

At the end of the article, the company concluded that in the future, all images, as well as some text and algorithms, will be generated using AI.

Will humans lose their jobs because of AI?

It is foreseeable that with the explosion of generative artificial intelligence, more and more advanced AI programming models will appear one after another in the future, squeezing the living space of programmers.

So, will human programmers become unemployed due to the development of AI technology?

An industry consensus is that if they want to replace humans, "AI programmers" still have many problems to solve.

This is mainly reflected in two aspects: "commercialization prospects" and "regulatory ethics".

Joanne Chen, a partner at Foundation Capital and an early investor in Jasper, said it remains difficult to turn a generative AI tool into a valuable company.

Not long ago, Kite, the "AI programming artifact", announced that it would stop development, and it was completely inactive after only 8 years of operation.

In the last blog posted on Kite’s official website, founder Adam Smith said, “We have 500,000 monthly active users, but are generating almost no revenue.”

He believes that just making developers write code 18% faster is not sensational enough for them, nor is it enough for them to spend money on value-added services.

The commercialization prospects of the product are not clear enough, which may also be a common problem of many paid AI auxiliary software.

"Ethics and copyright" are another obstacle that prevents generative AI technology from entering people's lives.

Earlier this month, a class action lawsuit was filed against GitHub, accusing it of using the Copilot tool to use artificial intelligence to copy open source code and disregarding software privacy.

Some developers also complained that the code suggested by Copilot looked like their own work.

While GitHub says that in rare cases the tool can generate copied code, the current version attempts to filter and block suggestions that match existing code in GitHub public repositories. But it's still generating considerable anxiety among some in the programmer community.

Chen also said the boom in generative AI could mean a lack of regulation and allow them to be used for "some unsavory or dangerous purposes." Such as making videos that spread misinformation.

So in terms of Google’s possible launch of Pitchfork, while the project is still in its early stages, there are still thorny ethical issues to consider on how to train these models, such as bias and potential copyright issues. .

So, are programmers "killing" themselves?

The above is the detailed content of Google’s mysterious project exposed! An AI that can write code and fix bugs makes coders tremble. For more information, please follow other related articles on the PHP Chinese website!