Altman promised to "evolve into a PhD"... OpenAI's ambitious work shamed

OpenAI's 'ambitious project' GPT-5 repeatedly makes basic mistakes

Users criticize the results on social media for being inaccurate

U.S. network CNN reported on the 14th (local time) that OpenAI's new artificial intelligence (AI) model GPT-5 is becoming a laughingstock among users due to performance that falls far short of expectations.

OpenAI unveiled GPT-5 on the 7th. When the previous models, GPT-3 and 4o, were first released, they completely changed the landscape of the AI ecosystem, so the industry and users were highly enthusiastic.

Sam Altman, OpenAI CEO, had promoted the model as "at the level of a PhD expert." During an online briefing on the 6th, CEO Altman said, "If GPT-3 felt like talking to a high school student, and GPT-4 felt like talking to a college student, then GPT-5 should feel like conversing with a PhD-level expert." Altman also evaluated GPT-5 by mentioning Artificial General Intelligence (AGI), referring to AI with intelligence equal to or greater than a human, saying it was "a big leap forward and a significant step toward AGI."

The biggest change in GPT-5 is that users no longer need to select a model—the AI itself chooses the most suitable response.

However, actual users of GPT-5 have been shocked by the frequent errors and clearly incorrect answers. In response, users have started to post botched results produced by GPT-5 on social media, mocking its poor performance.

For instance, when prompted to create a table listing photos and names of the initial 12 U.S. presidents, the AI rendered 'George Washington' as 'Gearge Washingion' and 'Thomas Jefferson' as 'Thomason Jefferson' in the results.

Additionally, when asked to draw a map of the United States and label the states, it wrote 'California' as 'CALFORHIA' and 'Idaho' as '1OAHO.'

Some users were so disappointed with GPT-5—believing it even worse than the previous GPT-4o model—that they launched an online petition to bring back the prior version. OpenAI eventually complied.

On the 9th, CEO Altman posted an explanation on X (formerly Twitter), saying, "The 'auto-switching device' (in the model) had broken, making GPT-5 appear dumber than it really was. Starting today, it should look smarter." What happened was an automatic switching device, intended to select the most appropriate model in response to user questions, had failed early on.

Nevertheless, CNN pointed out that even after Altman's explanation, when prompted to draw a map of the United States, the AI's responses were still poor. CNN analyzed that user disappointment with GPT-5 was an inevitable result of OpenAI's excessive hype.

ChatGPT currently has about 700 million weekly users, making it the most widely used generative AI in the world.

Jungmin Oh, Hankyung.com reporter blooming@hankyung.com