Don't Ask For A Short Response To AI Chatbot, This Is What Will Happen

JAKARTA If you often use Artificial Intelligence (AI)-powered chatbots, both in the form of a web or application, you must realize that this technology often provides a long response.

This is indeed regulated by the system of the AI model used. The more sophisticated the AI model is, the more detailed the response is given. In fact, if chatbot gives a concise answer, this could be a problem.

Users shouldn't even ask for concise answers from chatbots. According to a new study by Gischard, a Paris AI testing company, a concise request for answers will have a negative impact on the AI capabilities they use.

Worse, this will cause hallucinations. The more frequent chatbots are hallucinatory, the worse the response that will be given. This means that the performance of this technology has decreased.

"Our data shows that a simple change in system instruction dramatically affects the model's tendency to hallucidate," the researchers continued, quoted by TechCrunch.

Hallucination is a major problem for this model because it is difficult to overcome. In fact, sophisticated models made by OpenAI and Meta can still imagine even though many people have used their chatbots.

When a model experiences hallucinations, the technology will make up a story. Based on researcher Gischard's explanation, requests regarding concise answers can confuse the system because it often contradicts its questions.

For example, when users ask for an explanation of the history of the second world war, they ask for a concise answer. These two things are very contradictory because history usually requires a very long explanation.

There are several AI models that Giskard tested with a concise answer order. Some of these models are GPT-4o from OpenAI, Claude 3.5 Sonnet from Anthropic, Deepsek V3, Llama 4 from Meta, to Grok 2. All of these AI models experienced a decrease in accuracy after being asked to answer concisely.