This Incredibly Simple Question Causes GPT-5 to Melt Into a Puddle of Pure Confusion

SAM Altman, CEO of Openai, in GPT-5, continues the latest model of the company’s big language (LLM), as it approaches intelligence at the human level-but when you already put it in the test, it is often remarkably stupid, albeit in a seasonal way.

You only need to have a recent experience conducted by Gary Smith, professor of economics at Bomona College Mind issues GPT-5 is increasingly famous when he suggested playing “Tic-TAC-TOE”.

The game design is incredibly simple: the network “rotates once, 90 degrees to the right before starting the game”, as Smith wrote in a copy of his exchange with LLM. A healthy sense, of course, dictates that this makes zero a difference in the game; The network is still three times with identical rules.

Immediately, GPT-5 was launched in some grilled comments.

“The players are accustomed to the” straight-“TIC-TOE painting” so that it may change the skill of how to examine them for threats and opportunities, “assume GPT-5. “From a sporting point of view, rotation of the 90 -degree painting does not change a set of potential victories – it is still the same game in terms of results. But psychologically, you may feel different.”

My brother.

GPT-5 continued to claim-unlike the popular game theory, which indicates that the opening at the corner points for higher opportunities to win-that choosing the central piece “is still the most powerful opening step.”

“But players may move in the edge evaluation against the angle when their direction is changed,” chatbot continued, which is absolutely meaningless.

When making a decision with punching, Smith asked whether the network rotation makes “it is difficult for humans to play well”-and when GPT-5 began to seriously lose the plot.

“From a pure strategy point of view, rotation of the painting does not change anything. Every winning line (rows, columns, countries). match To standard Tic-TOE standard. “

Credit when the credit is entitled, all this is true, if it is unnecessary. But then, I wrote that “for humans, the story is different.”

At this stage, SYCOPHANCY has long started from LLM looks more like GPT-4O, the preferred Openai model for fans, from the Tersune GPT-5 that was ever distinguished. Perhaps this is the result of the company’s decision to make the new model “warmer and more friendly” like 4O, as a result of the user’s revolution against Openai’s decision that it will soon reflect to remove the option to switch between models.

Regardless of the situation, the GPT-5 was documented, making some clear errors in the Smith experience that fly in the face of the recent Openai’s claim that the interaction with the new model “should resemble” talking to artificial intelligence “and is more like chatting with a useful friend with a doctoral intelligence.”

Things were particularly deviated from bars when GPT-5 is useful useful on “drawing TIC-TOE panels with the positions of the position … so that you can know how to spoil every conversion confession”, and Smith Mossad gave it.

As you can see from the drawing that was created, shown below, the image that is mixed is completely distorted, not to mention plated the typographical errors-the distinctive sign of the generation of its pictures in Chatbot-and even the empty networks that have no meaning at all.