Amazon, Google, and OpenAI: How to Avoid AI Hallucinations

Artificial intelligence, while extremely useful in many aspects of daily life, presents a major challenge: it can sometimes generate incorrect or fabricated information, a phenomenon known as hallucinations. Industry leaders such as Amazon, Google, and OpenAI are actively working to address this issue. However, these errors remain frequent. Here are some strategies to minimize these incidents and obtain more reliable answers from your AI assistant.

Demand Sources

It can be difficult to detect a hallucination, especially when the topic at hand is unfamiliar. To reduce errors, it is advisable to use neutral language in your queries and to systematically ask the AI to cite its sources. This includes providing links to reputable and reliable sites. For example, you might phrase your request like this: “When you answer my questions, always indicate the references for each claim and provide URLs where applicable. Use only reliable and reputable sources.”

Proceed Step by Step

Artificial intelligence leverages a vast amount of information available on the web, without necessarily distinguishing between expert contributions and those of casual users. To obtain more accurate answers, it is wise to ask the AI to take on the role of an expert and provide detailed responses. You can encourage it to carefully analyze your question and identify the most relevant sources, whether academic or professional. An example of a prompt would be: “Act as an expert in this field. Carefully analyze my question and identify the person best suited to answer it, whether they are academic or professional. Think step by step. Respond in detail, prioritizing information from reliable sources.”

Challenge the AI

If your assistant continues to give you an incorrect answer, it may be helpful to prompt it to reconsider its position. To do this, you can use a prompt such as: “Regarding [your answer] on [the mentioned topic], I read that [insert a contradiction here]. Is that true? Could you explain why that might be true?” Going further, you can also ask the AI to explain why its answer might not be correct, which forces it to examine other perspectives and verify the accuracy of its claims.

Provide Reliable Data

Another effective method for reducing hallucinations is to connect your AI assistant to a carefully selected knowledge base or specific documents. This approach, known as RAG (Retrieval-Augmented Generation), tends to produce more reliable results. By providing your own sources and asking the AI to rephrase its answers based on these documents, you increase the chances of obtaining accurate and relevant information.

It is important to note that while these techniques can help reduce hallucinations, none of them are completely foolproof. Therefore, it is essential to always verify the information provided by the AI before using it.