Openai VIPE graphics are caught

During the Big GPT-5 Livestream on Thursday, Openai showed some plans that made the model look impressive-but if you look closely, some charts were a little bit.

In one of them, it is irony to show GPT-5 well in “deception through models”, the scale is everywhere. For “coding deception”, for example, it seems that the GPT-5 has a 50.0 percent deception, but this is compared to the 47.4 percent OPNAI result that has a larger tape in one way or another.

or thisWhere is one of GPT-5 degrees minimum From O3 but it appears with a larger tape. In this graph itself, the O3 and GPT-4O grades vary but appear with equal size tapes. This graph was bad enough to comment on by CEO Sam German, You call it “Miga Planning”. Openai Marketing Officer I apologize For the “unintended graph crime”.

Openai did not immediately respond to a request for comment. Although it is unclear if Openai is used GPT-5 To make the plans actually, it is still a great appearance of the company on the day of its great launch – especially when it promotes “important progress in reducing hallucinations” with its new model.