Laid off? Programmers are temporarily safe

Estimated read time 10 min read

Before the emergence of ChatGPT, the idea of ​​“AI replacing humans” had always been considered a “science fiction” gimmick by AI companies. However, after seeing the super capabilities emerging from large language models, people really started to worry about their “jobs”, especially programmers.

In March 2024, the startup Cognition demonstrated Devin, one of the company’s AI programming tools. Cognition describes Devin as “the world’s first fully autonomous AI software engineer.”

In Cognition’s demonstration video, Devin can build an application from scratch by only accepting natural language instructions. Devin also completed an actual task order on the crowdsourcing platform Upwork, not only completing the code writing and debugging of the computer vision model, but also preparing a report at the end of the task.

Cognition even posted on the social media platform that Devin has passed the interview as an engineer at a leading AI company.

At this point, the saying “those who develop AI will be the first to be replaced by AI” is rapidly gaining popularity on the Internet.

However, just a few days ago, Carl, an American engineer who claimed to have 35 years of development experience, posted a video on his YouTube account Internet of Bugs, saying that Devin’s demonstration may be fraudulent.

Carl, an American engineer who claims to have 35 years of development experience, posted a video on YouTube, saying that Devin’s demonstration may be fraudulent.

Maybe Devin is just “playing” a programmer?

Currently, Devin is not open for use, and applications can only be submitted through email. Therefore, the outside world’s understanding of Devin basically comes from the official demonstration video and the evaluation of a few third-party developers and product personnel.

Among third-party reviews, the most popular ones include a report published by Bloomberg on March 12 about Cognition AI and several of its Chinese founders . This report mentioned that the reporter used Devin to build a website from scratch in 5-10 minutes. The report also quoted Silas Alberti, a computer scientist who had tried Devin, as having a high evaluation of Devin.

Although Carl, the blogger who questioned the cost of Devin, did not actually test Devin. However, by analyzing the Devin practical video released by Cognition AI, he proposed many suspected fake contents in the demonstration, including:

1. In the demonstration video, Devin did not actually understand or correctly complete many of the operations. The report generated by Devin did not contain the information the client actually needed.

2. When fixing the error code, Devin also encountered the problem of ” doing it even if there is no possibility of creating difficulties” . Carl pointed out that some of the code Devin fixed were mistakes that “human beings would never make.” Therefore, it is possible that these errors do not come from the customer’s code base, but are caused by Devin himself.

3. Technically, Devin’s performance was not very good either. Sometimes outdated methods of handling files are adopted , inappropriate operations are performed within the Python environment, or even writing one’s own low-level file reading loops instead of using the standard library correctly.

4. The video looks like Devin completed the task very quickly, and the video creator was able to complete the requested task in about 30 minutes, but the timestamp in the chat shows that the task lasted for multiple hours and even continued into the next day. .

Although Carl believes that Cognition has exaggerated in Devin’s demonstration and actual operation, he also said that Devin has shown impressive capabilities in certain technical operations. For example: Devin is able to automatically update and configure files to match the required library versions. He believes that with the advancement of technology, AI tools like Devin will be able to fully understand and perform more complex programming tasks.

Are humans still the key to AI programming?

At present, Devin has not been launched yet. It is difficult to tell whether its capabilities are really as strong as Demo.

But what is certain is that although AI can indeed help humans program, the mainstream view still does not believe that AI can replace humans .

“AI is far not as smart as outside rumors say. To understand business logic requires a lot of repeated communication, let alone work.” A developer who pays attention to AI programming told Huxiu that many programmers are currently using Copilot. tool, but not many people are worried about being quickly replaced.

In fact, there are still two difficult problems to solve in the current work of large-scale language models. On the one hand, the input information may not necessarily correctly understand the underlying logic, and may require repeated communication with AI. On the other hand, it is difficult to avoid errors and illusions in the content output by AI. In work that requires more rigor and seriousness, there will be great risks without human inspection and supervision.

“When human programmers write programs, the results are basically certain, either good or bad. But after the big model comes, it’s hard to say.” Alipay CTO Chen Liang said that there is great uncertainty in the results of AI programs . “He can write better than you, or he can write worse than you.” This requires human programmers to find ways to control it or influence it.

Chen Liang believes that although the output results cannot be determined like writing programs before, he can be influenced through translation, corpus, data, etc., so that AI has a greater probability of providing good results.

When Cognition just released the Devin demonstration video, former Tesla AI technical director and OpenAI co-creator Andrej Karpathy once commented on X:

“In my opinion, automated software engineering looks similar to autonomous driving. The development of autonomous driving is:

1. First humans perform all driving actions manually

2. Then the AI ​​helps keep the lane

3. Slow down when you see the car in front of you

4. It also changes lanes

5. It will also stop at signs/traffic lights and take turns passing

6. Eventually you will get a fully functional solution and continue to improve the quality until fully autonomous driving is achieved.

In software engineering, progress works similarly. Both AI does more and humans do less, but still needs supervision:

1. First humans write the code manually

2. Then GitHub Copilot automatically completes a few lines

3. ChatGPT writing code block

4. Move to larger and larger code differences (e.g. Cursor copilot++ style)


Andrej Karpathy believes that Devin is an impressive demonstration that may promote the advancement of coding tools, including terminals, browsers, code editors, etc., as well as human supervision.

What is the AI ​​crisis for programmers?

Although Devin may not be able to replace humans, many human jobs are indeed threatened by AI.

“AI does not necessarily replace people, but people who can use AI will definitely replace people who can’t use AI.”

Since 2023, the importance of AI tools to work has risen to a new level.

Chen Liang said that currently 60% of Ant programmers use Codefuse when writing code. About 12% of the code written by Ant programmers using Codefuse on the entire site is generated by Codefuse.

Yibo, the founder of Yichuang Technology, told a more vivid example.

“At the beginning of 2023, I required all programmers in the company to be able to write code using ChatGPT. They took a month to study and assess. After the assessment, the average efficiency increased by 3-5 times . The programming efficiency of one of our strongest architects is basically the original 5 to 10 times.”

Yibo said that before using Copilot tools, the architect must first sort out the customer’s needs, and then arrange for 5 to 10 people to help him work. He must tell these people: You are responsible for the front-end framework. You write the backend interface.

“There is no need to do this now. After the architect has thought it through now and written it clearly in the document, he can directly use ChatGPT to get it out.” It may take a team of 10 people to explain one thing clearly, do it well, and check it clearly. one week. Now this may take less than a day, or even just a few hours to complete.

Devin is innocent, guilty of hype

Devin brought his creator Cognition to the forefront. This company, established at the end of 2023, has received approximately US$21 million in investment from well-known institutions such as Founders Fund. The company’s seed round valuation was approximately US$350 million, and its current valuation has reached US$2 billion.

Now, as Devin’s title of “the world’s first AI software engineer” is questioned by Carl as false propaganda, Cognition may be facing a public relations crisis.

Regarding “false propaganda”, Carl criticized that the promotion of AI products often lacks transparency and does not truthfully display the actual capabilities and limitations of the product, thus misleading the public and professionals.

In the past year, there have indeed been many cases of “false propaganda” that have exploded in popularity of AI.

The most famous one is “Google releases Gemini demo video”. Google demonstrated the multi-modal capabilities of experience in a demonstration video released by its latest large AI model Gemini. But the company later admitted that the video was actually edited, using static images and text cues rather than real-time motion processing. Reduced latency in video, simplifying Gemini output.

Not long ago, Amazon launched a “no checkout” technology (Just Walk Out), which was also pointed out as artificial “disguised” artificial intelligence. Reports indicate that this technology does not entirely rely on advanced AI and computer vision, but rather involves more than 1,000 Indian staff manually reviewing customers’ shopping behavior to ensure checkout accuracy. However, Amazon has now denied this statement and claimed that these staff are mainly training models and helping to improve the technology.

In an earlier case, a well-known domestic AI company was accused of disguising the content of human simultaneous interpretation as AI translation content at a conference and displaying the company’s product logo.

In the video accusing Devin of fraud, Carl called on manufacturers, media and users to be skeptical of AI technology and conduct necessary verifications. Carl believes that this kind of exaggeration not only misleads the public, but also sets unrealistic expectations for AI technology.

Some netizens also expressed support for Carl’s views.

rayhere7925 believes: Many of these technology startups have only one goal: to make money. How do they get funding? Through investors and/or hype. I’m sure this over-the-top presentation, and all the articles and influencer videos that followed it, must have convinced an investor somewhere to put money into this company – which is exactly what this company is aiming for. Companies will lie if it means making more money (if they can get away with it). It’s no surprise that this company lied to create hype and attract investors.

Apexphp believes: The problem with modern journalism is that people who wrote technical articles in the past usually had some knowledge of the field. They knew the meaning of the professional terms and were able to read scientific papers in the field in which they were writing and understand most of them. But now we just have a bunch of pay-per-article people who find interesting headlines on the internet and simply write a few sentences because the company they work for pays them as much as they do for a well-written, well-researched article They pay the same fee, so there’s no need for them to bother.

You May Also Like

More From Author

+ There are no comments

Add yours