Controlling the mental effort of the model
OpenAI has launched a new feature that allows users to determine the level of mental effort a model exerts when generating answers, with the goal of improving analysis accuracy and response depth. This control allows the model's performance to be tailored to the nature of the task, whether it's simple questions that need a quick answer or complex tasks that require deeper thinking.
Different Effort Levels
Effort levels range from Light for quick and simple answers, Standard for typical daily use, Extended for moderately complex tasks, and Max, known as "Juice 200," which provides the maximum level of focus and analysis. Each of these levels affects how the model processes information and presents results more accurately or in more detail.
Limitations and feedback
The feature is currently only available for Pro and Enterprise plan users, due to the high computing requirements. However, some details are still unclear, such as the impact of increased effort on model speed or resource consumption, the maximum power consumption limits are not disclosed, and there is no information on whether the feature will be available in other plans in the future.
Risks of overthinking
Studies have shown that raising the mental effort level of the model can sometimes lead to "overthinking", which can negatively affect the accuracy of the answers in some cases. This suggests that the level of effort should be balanced according to the nature of the question to ensure the best results without slowing down performance or confusing the model.
Branching from here
In addition, OpenAI introduced the "Branch from Here" feature that allows creating new paths for the conversation from a given point, making it easier to test multiple scenarios and compare model results. This feature is useful for advanced users, but details of its impact on data storage or the maximum number of branches are not yet revealed, and there is no information on whether unnecessary branches can be deleted or made available to other users in the future.
OpenAI enables control of model effort for smarter and deeper answers.
