GPT-5 Introduces Universal Verifier to Combat AI Hallucinations — The Information

The publication The Information has released an extensive article detailing the development of GPT-5, shedding light on various technical features of the anticipated OpenAI model. This new version will utilize reinforcement learning technology and incorporate a Universal Verifier—a specialized module designed to cross-check every response generated by the model. This addition is expected to significantly reduce the incidence of hallucinations.

According to The Information, the development of this new model leveraged insights gained from GPT-1 and GPT-3, which excel at complex tasks but are less effective as chatbots. In contrast, GPT-5 is well-suited for casual conversations, while its capability in handling complex problems has been elevated; it is even suggested that it can be entrusted with refund tasks in customer support. When it comes to programming challenges, the model is said to generate cleaner, user-friendly, and feature-rich code.

Other enhancements include reduced computational resource consumption per token and improved agent capabilities. The publication also confirmed that GPT-5 will autonomously determine the amount of resources to allocate for a given task—rendering the current menu that allows users to choose among various models in the ChatGPT interface obsolete.

The development of GPT-5 has proven to be a challenging phase in OpenAI’s history. Initially, it was expected that GPT-5 would be released as Project Orion; however, it ultimately launched in February as GPT-4.5 and was quickly deemed a dead end due to its minimal performance increase over GPT-4. Currently, as reported by The Information, one of OpenAI’s challenges is the aggressive strategy employed by Meta* (recognized as extremist in Russia), which is investing heavily to lure away the company’s top researchers.

P.S. You can support me by subscribing to the channel «the runaway neural network«, where I explore the creative aspects of AI.