Sam Altman: OpenAI’s Approach to Addressing AI “Hallucinations” Aims for Better Explainable AI

Sam has already stated that he wants models to be more like reasoning engines than knowledge repositories, on the one hand. On the other hand, even in this situation, the model must be able to draw from a foundation (our history) and work with data.

OpenAI is making significant progress in addressing the issue of AI “hallucinations” by developing a novel AI model training method. Concerns about misinformation generated by AI systems, particularly in domains requiring complex reasoning, have prompted a focus on hallucination mitigation.

When models fabricate information and present it as factual data, AI hallucinations occur. OpenAI’s new strategy, known as “process supervision,” aims to address this issue by encouraging human-like thought processes within the models. The research aims to identify and correct logical errors or hallucinations as a first step toward developing aligned AI or artificial general intelligence. As part of this effort, OpenAI has released a comprehensive dataset consisting of 800,000 human labels that were utilized to train the model referenced in the research paper.

While the development of “process supervision” represents a promising advancement, some experts remain cautious. Senior counsel at the Electronic Privacy Information Center expressed skepticism, highlighting that the research alone does not fully alleviate concerns surrounding misinformation and inaccurate outcomes when AI models are deployed in real-world scenarios. To further evaluate the proposed strategy, OpenAI is likely to submit the research paper for peer review at an upcoming conference. As of now, OpenAI has not responded to requests for comment regarding the implementation timeline for integrating the new strategy into ChatGPT and other products.

OpenAI’s CEO, Sam Altman, emphasized the significance of striking a balance between creativity and accuracy within AI models. Altman envisions models that function as reasoning engines, not just repositories of knowledge. However, he also acknowledged the need for models to rely on a foundational base, such as historical data, and effectively process factual information.

The development of this innovative approach and the ongoing efforts to address AI hallucinations showcase OpenAI’s commitment to advancing the field of AI while ensuring responsible and reliable outcomes. As OpenAI continues to refine its strategies and seek solutions to the challenges posed by AI hallucinations, the prospect of achieving better explainable AI becomes increasingly tangible.

• OpenAI’s ChatGPT, a chatbot powered by the advanced GPT-3 and GPT-4 models, has witnessed unprecedented growth, surpassing 100 million monthly users in a record-breaking two months. With Microsoft’s substantial investment of over $13 billion, OpenAI’s value has soared to approximately $29 billion.

Read more about AI: