For all the hype, GPT‑5 is the first model that feels like it undoes hallucinations… Here it not only corrects its own work, it corrects *the user’s* work…

In the video above, we purposely threw GPT-5 a curveball: we asked it to build a model for a Norwegian company called “Autoshop”, which didn’t exist. After crawling CapIQ tables for real companies, some of which were closer to the prompt, GPT-5 inferred that the intended company was “AutoStore” based on location it pulled from the tool, and corrected the user.