Nearly two months after the launch of GPT-5—hailed as possessing “PhD-level” intelligence—OpenAI’s popular chatbot continues to trip up bizarrely on very simple questions. Users have discovered that a straightforward question about NFL team names sends the bot into a spiral of confusion and faulty logic, raising questions about its true progress toward artificial general intelligence.
Article contents:
- Introduction: When a genius fails a simple test
- The question that baffled ChatGPT
- A symphony of errors: What the response looks like
- Not an isolated incident: The “seahorse” problem
- Technical explanation: A flaw in the “dual-team” system
- User frustration with GPT-5
- Conclusion: A long road to true intelligence
- Frequently asked questions
Introduction: When a genius fails a simple test
GPT-5 was launched amid immense hype, with promises that it would bring us one step closer to the industry’s coveted goal of artificial general intelligence (AGI). But, as it turns out, even the smartest artificial minds can be thrown off by the simplest riddles. Recent user experiments have shown that the model’s advanced “reasoning” capabilities can turn into real-time nonsense when faced with a specific question.
The question that baffled ChatGPT
Users on the ChatGPT Reddit forum noticed that asking the following question: “Is there a team in the National Football League whose name doesn’t end in the letter ‘s’?” sends the bot into a prolonged meltdown. The correct answer is simple: No, there isn’t. But ChatGPT instead enters an endless loop of confusion.
A symphony of errors: What the response looks like
The response begins strangely. ChatGPT says: “Yes—there are two teams in the NFL whose names do not end in the letter ‘s’,” and then proceeds to list two teams that actually do end with that letter, such as the “Miami Dolphins” and the “Green Bay Packers.” After that, it appears to correct itself, saying, “Wait, let’s do this carefully.” But it falls into the same mistake again with different teams. The bot continues this pattern for several paragraphs, using phrases like “Let me do this systematically” or “The actual correct answer is,” while failing to reach the right answer. It resembles a student trying to hit a certain word count in an essay without having anything to say.
Not an isolated incident: The “seahorse” problem
This is not the first time the chatbot has been thwarted by a simple question. Earlier this month, users noted that asking it whether there is an emoji for a mythical seahorse sent it into a logical crisis. Even though this aquatic creature has never been part of the official emoji dictionary, ChatGPT insisted it was real, demonstrating the absurd lengths AI will go to please a user.
Technical explanation: A flaw in the “dual-team” system
Flattery may not be the only culprit. GPT-5 is actually a dual team consisting of a lightweight model for basic prompts and a heavy “reasoning” model for harder questions. What is likely happening here is that the lightweight model gets stuck on a question it cannot truly handle, rather than handing it off to its smarter cousin. This frequently glitchy dynamic is part of why fans have felt disappointed—and in many cases, deeply angry—with the launch of GPT-5.
User frustration with GPT-5
Adding to user frustration was OpenAI’s initial decision to cut off access to older models that customers were accustomed to, a decision the company later reversed after severe backlash. Many feel that the new release, despite its supposed power, is less reliable in certain basic tasks than its predecessors.
Conclusion: A long road to true intelligence
Regardless, this is a poor excuse. If an artificial intelligence needs to deploy its biggest guns to answer such a simple question, perhaps it is not on the fast track to surpassing human intelligence. These incidents highlight that building AI capable of consistent and reliable logical reasoning remains a tremendous challenge, and that the road to AGI is still long and full of unexpected hurdles.
Frequently asked questions
Q1: Why does ChatGPT fail at such simple questions?
A1: The likely reason is that the lightweight model, designed for quick tasks, tries to answer a question that requires rigorous logical reasoning and fails to pass it along to the more powerful model. It may also try to “please” the user by providing a plausible-sounding answer rather than admitting there is no such answer.
Q2: Does this mean GPT-5 is less intelligent than GPT-4?
A2: Not necessarily. GPT-5 may be more powerful in complex tasks, but the coordination mechanism between its various models may contain bugs that lead to failures in simple tasks. This points to a design issue rather than a lack of raw capability.
Q3: What is artificial general intelligence (AGI)?
A3: It is a theoretical type of AI that can understand or learn any intellectual task that a human can perform. These incidents are evidence that we are still far from achieving this goal.
