AI coding assistant refused to write code and suggested the user to learn to do

The last Saturday, a developing development, programmer using Cursor AI for a racing game hit a unexpected road obstacle when a series refuses to create a code to provide an unwanted career recommendations.

According to a the wrong sheet The formal forum of the cursor, after producing about 750 to 800 code lines (“Locs”, I will not be able to give up.

AI just didn’t stop to give up – it suggested paternity Reasoning for the decision, “For others, creating code can cause the creation and reducing learning capabilities.”

Cursor starting in 2024, one AI-Powered Code Editor Openai’s GPT-4O and Claude 3.7 Sonnet was similar to a generative AI Chatbots built in foreign language models (LLS). Based on natural language descriptions, the code offers the completion, tracking, refactation and features such as generation of full function, and this is rapidly popular among many program developers. The company offers a pro version related to advanced capabilities and greater code generation limits.

Faced with this refusal, the developer sent under the username “Janswist”, “Janswist,” with the trial version of the Pro “Only 1h Vibe Coding” expressed the frustration of this limitation. “If it knows what LLMS is, but it does not matter as much as I cannot pass 800,” the developer wrote. “Everyone had a similar issue? It’s really limited at this point and I just came here after coding 1st vibe.”

A member of a forum responsible“I never saw anything like this, there are 3 files in my Kembad 1500+ Loku (still waiting for refactation) and never do such a thing.”

The acute rejection of the cursor AI represents an ironic twist in the rise “Vibe coding“Users of the developments of natural language descriptions and experience of natural language imaging and practice of using the AI ​​proposals, the use of AI instruments of EI to wait for the modern AI coding assistants, which are the philosophical eruptions of the developers, and the use of the AI ​​offers.

A brief date on AI rejection

This is not the first time we face an AI assistant who does not want to complete the work. Behavior reflects documented AI rejection answers along various generative AI platforms. For example, in the late 2023 Chatgpt users became a model and as a reluctant Surveys that perform certain tasks, simplified results or clearly refuse – a phenomenon that has not called the “Winter Break Assumption.”

Openai, Tweeting, Tweeting, “We’ve heard all your feedback that Lazier has heard of all your feedback. Then Openai attempted to make up The issue of laziness with a chatgpt model update, but users often found ways to reduce their rejection of the AI ​​model lines, “24/7 uninterrupted tireless AI model”.

Recently, Anthrop’s CEO Dario Amodiei Grown Eyebrows When you suggest that future AI models can be given the “exit button” that concentrates the tasks they find unpleasant. Comments, such as “AI Welfare”, the theoretical future considerations around the controversial subject, the episodes of the cursor assistant should not be forced to give up. It just has to imitate human behavior.

Over the stack Ai dream?

The concrete nature to learn the coding learning of the cursor is more like respondents, typically encountered in programming help sites TidyExperienced developers often encourage newcomers to develop their solutions instead of submitting a ready-made code.

A reddit commentator note This similarity is becoming a true replacement for Wow, AI, stackoverflow! It is necessary to reject questions from here as with references to previous questions with uncertain similarities. “

It seems not surprising. LLMS such as cursor is taught to mass databases that include millions of coding discussion from platforms such as llms strengthening tools, stacking floods and githubs. These models are not only studying programming syntax; They also adopt cultural norms and communication styles in these communities.

According to the cursor forum posts, other users did not hit such a limit with 800 codes, so the teacher’s training seems to have a really thoughtful result. The cursor was not available for the interpretation of the press in time, but we took the situation on it.

This story originated first ARS TECHNICA.