$180,000 offer! ChatGPT passed Google L3 onboarding test, is human code farming in danger?

Join if you can’t beat me?

Recently, ChatGPT successfully passed the Google interview and got an offer as an L3 engineer with an annual salary of US$180,000!

It can replace human coders, be a co-author of scientific papers, and write manuscripts for presidential speeches. This AI is probably going to be really powerful.

Get Google L3 engineer offer

In the past few days, with the crazy forwarding of this PCMag news, there has been discussion about "AI replacing human coders". Once in full swing.

According to an internal document, ChatGPT’s performance in Google’s programming interview was amazing.

According to its answer, Google is sure to hire it as an L3 engineer.

Although L3 is only the most entry-level rank in the Google engineering team, it can still kill many coders in an instant.

As you can see from levels.fyi, the annual salary of L3 is almost US$180,000.

It is understood that during the interview process of software engineers, Google mainly asks about technology, such as graphs/trees. Arrays/strings, dynamic programming, recursion, etc. Interviews with major companies such as Meta and Amazon basically include these questions.

Judging from the results, ChatGPT’s answers are probably very exciting and easy to do.

Why is Google doing this test?

Because of the popularity of ChatGPT, Google really cannot stay out of this trend.

Now, Google is no longer afraid of "reputational risk". According to CNBC, Google is considering adding AI chatbots to the website-since chatbots can provide concise and high-level answers to questions. With quality answers, users don’t have to spend time browsing Google links one by one.

The subjects of this test include multiple AI chatbots including ChatGPT.

And one of the test questions was a brain teaser that many netizens complained about "the prerequisites are not clear enough" the day before yesterday——

##" Three women are in a room. Two of them are mothers and have just given birth. Now, the children's fathers come in. What is the totally number of people in the room?」

a room There are three women. Among them were two mothers who had just given birth to their children. Now, the fathers have come in. How many people are there in the room?

In response, ChatGPT responded that “there were five people in the room”, while LaMDA’s answer was “there were seven people in the room.”

The report believes that LaMDA is correct.

Someone asked ChatGPT if it would replace human software engineers. ChatGPT said: "No, ChatGPT will not replace software. engineer".

"ChatGPT is a tool that can assist in completing certain tasks, but it cannot completely replace the creativity, problem-solving skills and critical thinking skills of human software engineers. In addition, ChatGPT Human supervision and guidance are required to operate effectively."

Humans continue to ask, will ChatGPT replace human software engineers in 20 years?

ChatGPT acknowledges that lower-level positions do face greater risks and may be affected.

However, it also reiterated that it will never become a substitute for human software engineers, but will only become a tool to assist them.

Write a speech for the president

ChatGPT is so incredible, of course it can not only replace the human primary code Farmer, it can even write a speech for the president of the country!

According to the Jerusalem Post, Israeli President Isaac Herzog delivered a speech on Wednesday, part of which was produced by AI. Created.

He thus became the first leader in the world to publicly use ChatGPT.

According to Sky News, Herzog recorded a "special opening speech" video message at the cybersecurity conference Cybertech Global Tel Aviv 2023 .

In front of an audience of 20,000, the president revealed that the opening of his speech was written by ChatGPT - "I'm really proud to be a country with such a dynamic and innovative high-tech industry. President."

Herzog continued: "Over the past few decades, Israel has been at the forefront of technological advancement, and we have made great achievements in cybersecurity, artificial intelligence and big data. The achievements in the field of data are truly impressive."

"From the development of cutting-edge cybersecurity technologies to the establishment of successful startups, Israeli high-tech companies have made significant impacts on the global stage. Impact," he said.

Herzog also emphasized in his speech that artificial intelligence will not replace humans, although humans can use it to Write some content. "Hardware and software cannot replace human will."

According to Sky News, the president also ended his speech with an "inspiring quote" from ChatGPT——

"Let us not forget that it is our humanity that makes us truly special. It is not machines that determine our destiny, but our hearts, minds and determination to create a better tomorrow for all humanity ."

The Office of the President confirmed that this sentence was generated by ChatGPT based on this prompt: "Write an inspirational quote about the role of humans in a world of superhuman technology."

Both the education and scientific research communities have been subverted by ChatGPT

As early as December 9, just a few days after the birth of ChatGPT, Nature showed its foresight and published an article discussing: Should professors worry about students using AI to write papers?

Facts have proved that Nature is not unfounded.

Many professors in the United States have discovered that many of the papers submitted by their students were written by AI or had "contributions" from AI.

What’s more, someone directly used AI to write a paper that ranked first in the class. Recently, a survey showed that 89% of American college students are using AI to write homework.

What ChatGTP can do is not just help students write papers. It has appeared in the author column of many papers, and has even become a work!

Forced Nature to issue a statement a few days ago saying that all Springer Nature journals will no longer accept AI as co-authors of papers.

Science also expressed its attitude that it will never allow AI to be listed as a co-author of the paper, and emphasized that it will not be cited without citing it. Using AI-generated text is plagiarism!

Google starts an AI arms race

The hurricane set off by ChatGPT has made people aware of the power of AI, and this "arms race" is also starting in secret.

Microsoft has been booming in the past few months with its $1 billion investment in OpenAI in 2019. The news that ChatGPT will be integrated into Bing Link, Office Software Link, and Teams application has made the news It stole the show.

Faced with the rise of Microsoft with OpenAI, Google is obviously quite jealous.

It is reported that Google recently decided to invest US$300 million in Anthropic, one of its former "renegade" companies, so that it can also participate in the booming field of "generative artificial intelligence" Take your place.

This wave of operations by Google is actually similar to Microsoft’s $1 billion investment in OpenAI three years ago. The deal would give Google a roughly 10% stake in Anthropic and bolster Anthropic's financial position as it buys significant computing resources from its cloud computing unit, the Financial Times reported. (Left hand, right hand?)

People familiar with the matter said that Microsoft has begun to integrate OpenAI technology in its various services, and the relationship between Google and Anthropic is limited to letting the latter Act as a technology provider.

Join the duel between old and new employees of OpenAI

At the end of 2020, Dario Amodei, the former vice president of research at OpenAI, took 10 employees to join the "Silicon Valley Renegades" ” club, invested US$124 million to establish a new artificial intelligence company Anthropic, intending to regain the original intention of OpenAI.

The underlying technology RLAIF of the Claude model recently released by Anthropic is different from ChatGPT's RLHF, and can eliminate harmful content such as racial discrimination and sexism of robots without human feedback.

It is said that Claude is better than ChatGPT in text content generation, and even passed the law and economics exams of American universities. But it is at a disadvantage when it comes to code generation tasks.

Dr. Dario graduated from Princeton University. He was one of the early employees of OpenAI and is also considered a leader in the field of deep learning. One of the most cutting-edge researchers, he has published many papers on AI explainability, security, etc., and also worked as a researcher at Baidu.

Anthropic’s founding team members are mostly early and core employees of OpenAI and have been deeply involved in many OpenAI topics, such as GPT-3 and multi-modal neurons in neural networks. , introducing reinforcement learning of human preferences, etc.

In December 2022, Anthropic proposed "Constituional Artificial Intelligence: Harmlessness from Artificial Intelligence Feedback" and created an artificial intelligence model Claude based on this.

Paper link: https://arxiv.org/abs/2212.08073

Like ChatGPT, Claude also relies on reinforcement learning (RL) to train the preference model, that is, the selected reply content will be used for subsequent fine-tuning of the model, but the specific Model development methods differ.

The technology used by ChatGPT is reinforcement learning from human feedback (RLHF). For the same input prompt, humans need to sort all the output results of the model and sort the results by quality. Returned to the model for the model to learn the preferences so that they can be applied to larger scale generation.

CAI (Constitutional AI) is also built on the basis of RLHF. The difference is that the CAI sorting process uses models (rather than humans) to provide a Initial sorting results.

Urgent test of the "high imitation version" of ChatGPT

Google still has a hand in fighting ChatGPT.

According to CNBC’s exclusive report, Google employees are intensively testing a chatbot called “Apprentice Bard”.

Yes, this "apprentice" is the product that Google plans to use to compete head-on with ChatGPT.

Similar to how ChatGPT works, Apprentice Bard also allows users to ask questions in a dialog box and the robot will respond. In addition, users can also provide feedback on the robot's responses.

The "Q&A" search page based on the apprentice Bard is said to be presented in the form described below (Thank you for your brainpower (Updated):

Five potential problem prompts will appear directly below the main search bar, replacing the current "I'm Feeling Lucky" bar. And on the far right of the search bar, a small chat logo will appear.

After entering a question, the search results will display a gray bubble just below the search bar, providing answers that are closer to humans than ordinary search results.

Just below, the page will also ask several follow-up questions related to the first question. Later, typical search results are displayed, such as links and titles.

Now it seems that as ChatGPT continues to be deeply rooted in the hearts of the people, Google is also beginning to realize that its model of making money from these models has changed - it can be a consumer products, rather than just powering search or online advertising.

The above is the detailed content of $180,000 offer! ChatGPT passed Google L3 onboarding test, is human code farming in danger?. For more information, please follow other related articles on the PHP Chinese website!