ChatGPT now fails at most basic tasks
Users report ChatGPT now consistently fails at simple logic, search, and identification tasks that competitors handle.
A growing chorus of users is reporting that OpenAI's ChatGPT has become noticeably worse at performing rudimentary tasks. Complaints include consistent failures in basic counting exercises, identifying images of common objects like hardware or handwritten text, and providing correct web links for simple search prompts. For historical research, the model now frequently misidentifies well-known events or people, confusing them with obscure alternatives. This perceived regression in core logical processing and factual reliability is striking as it occurs in areas where large language models (LLMs) were previously considered competent.
The decline is particularly frustrating for professionals who rely on ChatGPT for efficient information retrieval and basic reasoning. Users note that the model often provides blatantly wrong answers to simple A/B choices, only to issue a correction after being challenged—a pattern that undermines trust. This trend emerges while competitors like Anthropic's Claude 3.5 Sonnet and Google's Gemini are seen as making strides in general intelligence and consistency. While ChatGPT's coding capabilities may remain an exception, the overall dip in performance for everyday tasks is pushing some users to revert to traditional search methods for reliability, highlighting a potential quality control issue at scale.
- Fails basic logic tasks like counting and simple A/B comparisons, often requiring user correction.
- Struggles with image recognition for common objects and providing accurate web links for search prompts.
- Shows regression in historical fact-checking, confusing famous events with obscure ones, as competitors advance.
Why It Matters
If core reliability erodes, professionals may abandon AI tools for critical tasks, stalling productivity gains.