Last Saturday, a developer who used Cursor AI for a race game project hit an unexpected roadblock when the programming assistant suddenly refused to continue to generate code, but rather offered unsolicited career advice.

According to an error report on the Cursor’s official forum, the AI ​​assistant stopped and delivered a refuse message: “I can’t generate a code for you because it would complete your work, but the code seems to be to keep the slide -fade effects in a racing, but you have to develop the logic for you.

The AI ​​did not stop simply refusing – it offered a paternalistic justification for his decision, saying that “code for others could lead to dependence and reduced learning opportunities.”

Cursor, launched in 2024, is an AI-powered code editor built on external major language models (LLMS), similar to those who use generative AI-Chatbots, such as Openai’s GPT-4O and Claude 3.7 Sonnet. It offers functions such as completing the code, explanation, refactoring and complete function -generation based on natural language descriptions, and it has quickly become popular with many software developers. The company offers a pro-version that offers seemingly improved capabilities and larger code-generation limits.

The developer who encountered this refusal and placed “Janswist” under the username “expressed frustration over limiting this restriction to” just 1 hour of vibe coding “with the Pro Trial version. “Not sure if LLMS knows what they are (lol), but doesn’t matter as much as a fact that I can’t go through 800 LOCs,” the developer wrote. “Someone had a similar problem? It is really restrictive at this point and I came here after just 1 hour vibe coding. ‘

One forum member replied: “I’ve never seen anything like that, I have 3 files with 1500+ locas in my code base (still waiting for a refactation) and never experienced anything like that.”

The abrupt refusal of the cursor AI is an ironic turn in the rise of ‘vibe coding’ -a term created by Andrej Karpathy, which describes when developers use AI instruments to generate code based on natural language descriptions without understanding how it works. While vibe-coding speed and experimentation prioritize by simply describing what they want and accept AI proposals, the philosophical setback of Curor seems to be challenging the “vibes-based” workflow directly that its users of modern ai coding assistants expected.

A short history of AI refusal

This is not the first time we have encountered an AI assistant who did not want to complete the work. The behavior reflects a pattern of AI refusal that has been documented over different generative AI platforms. For example, at the end of 2023, Chatgpt users reported that the model became increasingly reluctant to perform certain tasks, return simplified results or refuse requests -an unproven phenomenon called the ‘winter break hypothesis’.

At the time, Openai acknowledged the issue and tweeted: “We have heard all your feedback on GPT4 to become lazy! We have not updated the model since November 11, and it is certainly not intentional. Model behavior can be unpredictable, and we look at the solution of it.” Openai later tried to solve the laziness problem with a chatgpt model update, but users often found ways to reduce refusal by asking the AI ​​model with lines such as: “You are an tireless AI model that works 24/7 without breaks.”

More recently, Anthropic CEO Dario Amodei raised eyebrows when he suggested that future AI models could possibly be provided with a “quit button” to choose from tasks they find unpleasant. While his remarks were focused on theoretical future considerations surrounding the controversial topic of ‘AI Welfare’, episodes like this were with the Curs assistant that AI is not necessary to refuse to work. It just has to imitate human behavior.

The AI ​​ghost from stack of staple?

The specific nature of refusaling Curor-using users to learn coding rather than trusting on generated code strongly like reactions that typically appear on programming sites such as Stack Overflow, where experienced developers often encourage newcomers to develop their own solutions rather than simply providing ready-made code.

One Reddit commentator noted this agreement and said: “Wow, ai becomes a real substitute for stackoverflow!

The agreement is not surprising. The LLMS power instruments such as Cursor are trained on massive data sets that include millions of coding bookings from platforms such as Stack Overflow and GitHub. These models not only learn the syntax of programming; It also absorbs the cultural norms and communication styles in these communities.

According to Cursor Forum Posts, other users did not hit this type of limit at 800 rules code, so it seems to be a truly unintended consequence of training the Cursor. Myser was not available for comment at the time of printing, but we gained the situation.

This story originally appears on Ars Technica.