Is GPT-5 Really Smarter Than a College Student
364Recently, OpenAI released GPT-5, their newest big language model, and everyone was excited to see what it could do. They claimed it had “PhD-level intelligence,” but some users and experts are now questioning if that’s really true. Just over a month later, reports show that GPT-5 still makes a lot of mistakes and even invents facts, a problem known as hallucinating in AI talk.
GPT-5’s Falsehoods Raise Concerns
Some people who tested GPT-5 found it often gives wrong information without realizing it. For example, a Reddit user asked about the gross domestic product (GDP) of different countries. The AI responded with figures that were twice as high as the actual numbers. Poland’s GDP, for example, was listed as over two trillion dollars, while reliable sources say it’s closer to $979 billion. The user only caught these errors because they seemed so off, which made them wonder how many times they might have accepted incorrect info without checking.
OpenAI has said that GPT-5 produces fewer hallucinations than previous models. But critics argue that’s not the whole story. The reason AI hallucinates is partly due to how these models are trained and tested. They’re designed to try and give the best answer possible, even if they’re guessing. Unlike some other AI systems, like Anthropic’s Claude, GPT models aren’t trained to admit when they don’t know something. That can lead to confidently wrong answers, which is risky when people rely on AI for facts.
Simple Tests Show GPT-5 Still Gets Things Wrong
Experts and skeptics have run their own tests to see how well GPT-5 performs. One researcher, Gary Smith, did some experiments, including a game of tic-tac-toe and asking for financial advice. When asked to label parts of a possum in an image, GPT-5 got strange results. It identified the animal’s nose as its leg or its tail as its foot, which is clearly wrong. Even when Smith misspelled “possum” as “posse,” the AI’s answers became even more bizarre, showing it struggles to understand simple concepts properly.
In another test, Smith asked GPT-5 to generate an image of a “posse” with five labeled body parts. The AI’s picture was hilarious but mostly wrong. It showed cowboys with lines pointing to the wrong places, and some labels didn’t match the animal or scene asked for. This kind of mistake makes it clear that GPT-5 isn’t anywhere near the “PhD-level” intelligence some claimed.
The Reality of AI Hallucinations and What It Means for Users
These tests highlight an important point: AI models like GPT-5 still make many mistakes and can confidently share false information. Users should be cautious and double-check facts when using AI tools. Relying blindly on AI for accurate data can lead to misunderstandings or misinformation.
OpenAI admits that current evaluation methods don’t effectively discourage hallucinations. They say that models are trained to get answers right, but they’re often encouraged to guess when uncertain. This means that, without proper safeguards, the AI might just make things up, especially when it’s unsure.
The takeaway? AI models are powerful but imperfect. They can be helpful for creative tasks or quick ideas, but for serious research or factual information, it’s best to verify everything independently. As AI technology advances, it’s important for both developers and users to understand its limitations and work to improve its accuracy.
In the end, GPT-5 shows us that AI still has a long way to go before it can fully replace human judgment. Until then, a healthy dose of skepticism and fact-checking remains essential.
What do you think?
It is nice to know your opinion. Leave a comment.