2024-03-22 07:49:00
Discussions are in full swing about artificial intelligence in the field of software engineering. A central subject: its impact in the sector with, in particular, the key question of the possible scrapping of human developers. Devin from Cognition Labs is likely to rekindle the debate with his presentation: “the first fully autonomous software engineering AI. » Publications are multiplying about the latter and analysts are of the opinion that it could mark the end of the need for human coders. What is it in reality?
Devin is part of the new wave of artificial intelligence specialized in software development. His presentation follows that of the AI called Magic.dev and announced to the public as a “superhuman software engineer. » This project also obtained $100 million in funding from Nat Friedman (CEO of GitHub from 2018 to 2021) and Daniel Gross (co-founder of the Cue search engine that Apple acquired in 2013).
Magic.dev and Devin share a common denominator: a value proposition that requires a strong presentation; “superhuman software engineer” or even “first fully autonomous software engineering AI.” » The feedback on this subject nevertheless indicates that Devin, in his current form, appears more like a coding assistant. He makes mistakes, is only satisfactory in 13% of concrete IT development cases taken from platforms like Upwork, and needs human assistance. The own website of the company Cognition, which publishes this artificial intelligence, does not rely on the latter for its creation. The use of Devin for the creation of the Cognition website would have been proof of its use in production.
The proponents of tools like Magic.dev and Devin are aiming for a goal as old as the dawn of computing: putting human developers out of business. n 1959, the COBOL programming language was designed by software engineer Grace Hopper. The stated goal of this language was to allow business people with no programming background to use it.
COBOL was announced as the answer to the increasing cost of computer development operations, including the salaries of computer developers: it allowed anyone, without programming experience, to write programs. The language was indeed friendly to professional users and was the most popular programming language in the world for a short time in 1970.
Ultimately, COBOL did not eliminate the need for developers: on the contrary, it created a demand for IT developers with specialization in this language. The current developments linked to the rise in power of artificial intelligence and its impact on the software development sector thus appear to be a kind of repeat of the COBOL episode.
Is Devin the AI that will replace software engineers? #ai #devin #coding pic.twitter.com/yif0N5rRAE
— Mosh (@moshamedani) March 16, 2024
Recent study concludes that generative AI will not replace developers anytime soon
Researchers at Princeton University developed an evaluation framework based on nearly 2,300 common software engineering problems assembled from bug reports and feature requests submitted to GitHub to test the performance of various large language models (LLM).
The researchers provided different language models with the problem to be solved and the code for the repository. They then asked the model to produce a feasible fix. This was then tested to ensure it was correct. But the LLM only generated an effective solution in 4% of cases.
Their specially trained model, SWE-Llama, was only able to solve the simplest engineering problems presented on GitHub, while traditional LLMs such as Anthropic’s Claude 2 and OpenAI’s GPT-4 were unable to. resolve only 4.8% and 1.7% of problems, respectively.
And the research team concludes: “software engineering is not simple in practice. Fixing a bug may require navigating a large repository, understanding the interaction between functions in different files, or spotting a small error in convoluted code. This goes way beyond code completion tasks. »
This is why Linux Torvalds wanted to distance itself from all the media hype around artificial intelligence. He considers it as a tool at the current stage of its evolution. He also suggests code review as an area of application of artificial intelligence. The ability of artificial intelligence to “guess” the developer’s intention will be useful to obtain reliable code in a short time. However, one condition will remain necessary: the developer must in turn examine what artificial intelligence offers him.
However, some speakers are of the opinion that current AI models only scratch the surface of the future picture of the software development industry and reinforce the idea of the possible disappearance of the developer profession
“ChatGPT and other GitHub Copilot just give an introductory overview of what the IT industry will be like in the future,” says the CEO of AI startup Fixie AI.
“Programming wizards like CoPilot only scratch the surface of what I’m describing. It seems quite obvious to me that in the future, all programs will be written by artificial intelligences, with humans relegated to, at best, a supervisory role. Anyone who doubts this prediction need only look at the very rapid progress being made in other aspects of artificial intelligence content generation, such as image generation. The difference in quality and complexity between DALL-E v1 and DALL-E v2 – announced only 15 months later – is staggering. If I’ve learned anything over the past few years working in AI, it’s that it’s very easy to underestimate the power of larger and larger AI models. Things that seemed like science fiction just a few months ago are quickly becoming reality.
I’m not just talking about Github’s CoPilot replacing programmers. I’m talking about replacing the very concept of writing programs with dedicated artificial intelligence agents. In the future, computer science students won’t need to learn skills as mundane as adding a node to a binary tree or coding in C++. This type of teaching will be outdated, like teaching engineering students to use a slide rule,” he adds.
And you ?
Do current developments in the software engineering field give rise to legitimate concerns about the future of human computer scientists in the field?
How do you see artificial intelligence in 5 to 10 years? As a tool or as a danger for your position as a developer?
See as well :
Is autonomous driving only a futuristic vision at Tesla Motors to date? The company has just changed the objectives of its Autopilot
SEC asks Musk to step down as Tesla chairman, demands US$40 million fine for out-of-court settlement
Tesla announces that the new computer for fully autonomous driving of its vehicles is in production and will be proven this month
Tesla shares fall after its autopilot system is involved in a crash and reports of its vehicle batteries catching fire
1711144521
#spell #human #coders #analysts