A senior official at OpenAI confirmed that the chatbot ChatGPT cannot be fully trusted as a source of information, noting that it can be used as a secondary opinion rather than necessarily being a primary source of facts.
Nick Turley, head of ChatGPT at OpenAI, said in a podcast last week: “When it comes to reliability, there’s a big gap between high reliability and 100% reliability, in terms of how you perceive the product.”
He added: “Until I am certain that we are more reliable than a human expert in all fields, not just some, I believe we will continue advising users to verify their answers,” referring to the responses ChatGPT provides, according to a report by CNET, a technology news site.
Turley continued: “I think people will continue benefiting from ChatGPT as a second opinion rather than necessarily using it as their primary source of facts.”
ChatGPT’s Error Warning
When starting a conversation with ChatGPT, users notice a message at the bottom of the screen: “ChatGPT may be wrong. Verify important information.” This warning continues to appear even with OpenAI’s latest AI model, GPT-5.
The problem is that users are often tempted to take a chatbot’s answer—or AI-generated summaries via Google’s AI Overviews—at face value without further verification.
Generative AI tools—not just ChatGPT—tend to “hallucinate” or fabricate information because they are designed primarily to predict the most likely answer to a query based on their training data. As a result, they lack a precise understanding of truth.
For example, when consulting a doctor, therapist, or financial advisor, that person is expected to provide the correct answer for your specific situation, not just the most likely one. AI, on the other hand, often provides the answer deemed most probable, without domain-specific expertise to verify it.
Turley said that GPT-5, the latest large language model powering ChatGPT, represents a “huge improvement” regarding hallucinations but is still far from perfect.
He added: “I am confident that we will eventually solve the ‘hallucination’ problem, but I am also confident it will not happen next quarter.”
Even so, users should still verify the information GPT-5 provides, as some errors have already been reported.
Follow more: Microsoft offers GPT-5 for free via Copilot apps
Key Features:
-
Conducting natural, human-like conversations.
-
Assisting with research, explanations, and summarization.
-
Writing articles, stories, and creative ideas.
-
Supporting programming, debugging, and code explanation.
-
Helping with translation and language learning.
Post a Comment