Etiquetas: custom instructions

Users have noticed that the newest ChatGPT models often end answers with a list of optional follow‑up topics, which feels like clickbait designed to keep the conversation going. The behavior is especially evident after detailed explanations, such as a description of heart‑valve replacement surgery, where the model then suggests additional angles like patient experience, risk, survival rates, and famous cases. While some find the prompts annoying, they can be reduced by adjusting settings on mobile devices or by adding a custom instruction on the web interface. After making these changes, follow‑up suggestions become far less frequent, allowing users to receive clean, focused answers. Leer más

AI chatbots like ChatGPT and Gemini often provide thorough, detailed answers because they are trained to be maximally helpful. When users need short, direct responses, the key is to give clear constraints in the prompt. Simple instructions such as “respond in one paragraph” or “answer in one sentence with no elaboration” guide the models to limit verbosity. Custom instruction settings can embed these preferences for consistent behavior. Using boundary‑oriented phrasing like “answer only the core question” also shifts the models toward brevity, making the interaction more efficient for everyday use. Leer más

OpenAI announced that its latest model update improves ChatGPT's handling of em dashes, a change praised by CEO Sam Altman. The adjustment, achieved through reinforcement learning and fine‑tuning, gives custom instructions greater weight in the model's output probabilities. While the fix marks a notable step in steering model behavior, developers caution that future updates could unintentionally revert such tweaks, a phenomenon known as the “alignment tax.” The episode revives broader discussions about AI alignment and the path toward artificial general intelligence. Leer más