Shulou(Shulou.com)11/24 Report--

Stability AI CEO said harshly: the "life span" of human programmers is only 5 years?

Recently, the founder and CEO Emad Mostaque of Stability AI once again said something amazing--

"within five years, human programmers will disappear completely. "

He said that he can foresee the future picture created by AI, and for human programmers, the future is really not optimistic.

AI code generation? As a matter of fact, Mostaque is not talking nonsense.

According to the data on GitHub, 41% of all code is currently generated by AI.

For example, GitHub's Copilot is the killer of AI programming.

Look at this effect, it's really one-click generation.

Even AI has become popular in cryptocurrency circles, and its popularity has soared.

Mostaque also said that Stability AI is also becoming more and more popular, and within three months, the popularity of GitHub has surpassed that of Bitcoin and Ethernet Fong, confirming this phenomenon.

And the territory of Stability AI is by no means limited to this.

They all dabble in the field of biology, DNA analysis, large language models, audio-visual data processing, and so on.

Looking to the future, Mostaque says there will be a breakthrough in the way people communicate and obtain information.

Starting with ChatGPT, Mostaque thinks that next year everyone will have one mobile phone, and they won't have to connect to the Internet.

Now these subversive AI models will be available everywhere in the future, and the way of dialogue and interaction will change fundamentally.

The next development of Stability AI is the construction of personalized model.

This is no small challenge, and the aim is to enable everyone, every company, and every culture to have their own AI system.

Mostaque explained that this is to achieve a key strategy of Stability AI-decentralization. This is much like the spirit of cryptocurrency, ensuring that AI technology is good for everyone to use.

At the same time, Mostaque believes that the future of AI development must be open and transparent, and technology should be fully democratized.

As we can see, Stability AI has opened up their LLM and downloaded it more than 2500 times.

Stability AI's model starts with 3B and 7B parameters, followed by versions of 15B to 65B. And they also released a RLHF fine-tuning model for research.

Performance thieves can fight. In fact, we are no stranger to AI code generation.

At that time, GPT-4 released and completely put the AI-generated code in the public eye.

The new generation of Copilot X directly integrates a gorgeous chat interface that can not only generate code with your mouth, but also talk to AI while writing code, as well as smart document systems tailor-made for you, and so on.

Specifically, the experiences supported by Copilot X are: Copilot Chat;Copilot for Pull Request;Copilot for Docs;Copilot for CLI. These new features are driven by OpenAI's GPT-4.

It is worth noting that due to speed delays, the code completion tool is still based on GitHub's Codex model, which is trained on GPT-3.

At that time, senior big data architect Zhu William said that Everything powered by AI is no longer far away.

Copilot can not only identify the code content entered by the developer, display the error message, but also deeply analyze and explain the use of the code block and generate unit tests. You can even give debug's advice.

In addition, in Copilot, you don't even need a keyboard to write code. Just sit in front of the computer and say, "Hey, GitHub!" "it can be realized by coding with open mouth. Currently, GitHub is experimenting with new features of GitHub Copilot Voice, a voice-based interactive system.

In addition to writing code, you can even do it through Copilot Voice: code jump, control IDE, code summary.

Not only that, some time ago, a tweet from Matthias Plappert sparked a wide discussion in the LLMs circle.

Plappert is a well-known computer scientist who posted the results of his benchmark tests on LLM, the mainstream of the AI circle, on HumanEval.

His tests are biased towards code generation.

The result is not very shocking, but also shocking.

Unsurprisingly, GPT-4 undoubtedly topped the list and won the first place.

Unexpectedly, OpenAI's text-davinci-003 sprang up and took second place.

Plappert says text-davinci-003 is a "treasure" model.

The familiar LLaMA is not good at code generation.

Of course, that's what AI can do now.

After all, the time limit given by Mostaque's bold words is five years.

Judging from the current progress, it is really not difficult to finish all AI programmers in 5 years.

Reference:

Https://decrypt.co/147191/no-human-programmers-five-years-ai-stability-ceo

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.