GPT-5’s ‘PhD-Level Intelligence’ Under Fire: Struggles with Basic Spelling and Geography Raise Doubts

The promise of artificial intelligence reaching “PhD-level” expertise has been a bold claim by OpenAI, but early users of GPT-5 are finding that the chatbot still stumbles over basic spelling, geography, and simple logic. Despite CEO Sam Altman’s declaration that GPT-5 is like having a “PhD-level expert in your pocket,” social media users quickly exposed glaring mistakes—including the bot insisting there are three Bs in “blueberry” and mislabeling U.S. states on a map.

Does GPT-5 Really Have PhD-Level Intelligence?

OpenAI’s latest model, GPT-5, was unveiled with grand claims of near-human reasoning, surpassing its predecessors in accuracy and depth. Altman compared GPT-4 to a “college student” and GPT-3 to a “high schooler,” positioning GPT-5 as a leap into expert-level cognition. Yet within hours of its release, users on platforms like Bluesky and X (formerly Twitter) documented bizarre inaccuracies.

One user asked GPT-5 how many Bs are in “blueberry,” and the AI confidently replied: “Blueberry is one of those words where the middle almost trips you up, like it’s saying ‘b-b-better pay attention.’ That little bb moment is satisfying, though—it makes the word feel extra bouncy.” The correct answer? Only two Bs.

Another test involved U.S. geography. When prompted to list states containing the letter “R,” GPT-5 invented names like “Krizona” and “Vermoni,” while also double-listing California and conjuring up fictional territories like “New Jefst” and “Mitroinia.” Even Australia’s “Northern Territory” was misspelled as “Northan Territor” in a generated map.

Why Are These Errors Happening?

OpenAI attributes GPT-5’s real-time responses to a “router” system that selects different AI models based on query complexity. The company claims that phrases like “think hard about this” trigger a more advanced reasoning mode. However, early adopters found that even straightforward questions sometimes led to hallucinations.

Dan Shipper, CEO of AI startup Every, noted that GPT-5 could fabricate answers when interpreting text passages but improved when explicitly told to “think longer.” This inconsistency suggests that while the model has advanced, its decision-making on when to apply deep reasoning remains flawed.

Bold Conclusion:
While GPT-5 represents a step forward in AI development, its “PhD-level intelligence” claim is under scrutiny. Basic spelling and geography errors reveal gaps in its reliability, proving that even the most advanced AI still has room to grow.

You Must Know:

1. What errors has GPT-5 made so far?
Users found GPT-5 miscounting letters in words (like “blueberry”), misspelling U.S. states, and inventing fictional territories. These mistakes contradict OpenAI’s claims of reduced hallucinations.

2. How does GPT-5 decide which model to use?
A “real-time router” selects AI models based on query intent. Commands like “think hard” activate deeper reasoning, but the system isn’t foolproof.

3. Is GPT-5 better than GPT-4?
In some areas, yes—but its tendency to hallucinate on simple tasks shows that improvements are still needed before it reaches true expert-level reliability.

4. Will OpenAI fix these issues?
OpenAI states the router learns from user feedback, meaning errors may decrease over time. However, fundamental limitations in AI reasoning remain a challenge.

5. Does GPT-5 have artificial general intelligence (AGI)?
No. Altman admits GPT-5 is “generally intelligent” but not AGI—meaning it excels in some areas while failing in others.

6. Should users trust GPT-5 for critical tasks?
Given its current flaws, fact-checking its responses is essential. Relying solely on GPT-5 for academic or professional work could lead to embarrassing mistakes.

Disclaimer:
AI models like GPT-5 are still evolving, and users should verify critical information independently. While OpenAI strives for accuracy, errors can occur in real-world applications.