AI models are getting smarter, but the cost is also climbing. According to foreign reports, some companies have recently started using a new prompt technique that requires AI to intentionally “speak less” when answering—even responding in brief sentences like primitive humans—all to reduce AI computing costs. The reason is simple: the longer the AI response, the more tokens consumed, and the higher the fees companies have to pay. So some companies are telling AI to stop giving long-winded responses and instead switch to “key points, brief, save where possible.”
Do Companies Really Care If AI Is “Dumb”?
Actually, companies have never cared about whether AI is dumb or not—they care about “how much money.” Taking OpenAI’s o3 model as an example, according to Anysphere’s testing, cutting response length by 50% reduces costs by about 66%. That’s why more companies are starting to require prompt engineers to optimize their prompts, making AI as “lazy” as possible—translated into plain English: “do less work, charge less.” Some prompt engineers even share various “slacking” techniques on social media, such as instructing AI to “respond with only three words” or “don’t explain your answers,” to achieve cost-saving goals.
Saving Money, but at the Cost of Quality
While the cost-saving effects are significant, this approach has also drawn criticism. Experts in the AI field warn that overly streamlining AI output may compromise the accuracy and depth of responses. When AI is asked to “speak less,” it may skip important reasoning processes and directly give “unexplainable” conclusions. Worse, this “lazy” mode could make AI more prone to hallucinations (Hallucination), because it lacks sufficient token space for complete reasoning and verification.