Full Report
OpenAI is working on a new feature called the Thinking effort picker for ChatGPT. [...]
Analysis Summary
# Main Topic
The ongoing development and testing by OpenAI of a new feature for ChatGPT called the "Thinking effort picker," designed to allow users to control the computational intensity ("thinking effort") the model applies to a given query.
## Key Points
- **Purpose:** To provide users with granular control over how much computational resource (or "juice") ChatGPT dedicates to generating a response, balancing response quality against speed.
- **Intensity Levels:** Four distinct levels are currently being tested with associated internal "juice" budgets:
- Light (Internal Attribute: 5)
- Standard (Internal Attribute: 18)
- Extended (Internal Attribute: 48)
- Max (Internal Attribute: 200)
- **Trade-off:** Higher thinking levels require more computational steps, generally resulting in deeper answers but significantly slower response times.
- **Access Restriction:** The highest level, Max thinking (200 "juice" budget), is currently gated and reserved for users on the $200 subscription (Pro plan).
- **Use Cases:** Light effort is suggested for quick questions, Standard for everyday use, Extended for medium complexity tasks (like econometrics or bond valuation), and Max for the most careful analysis.
## Threat Actors
- No specific malicious threat actors or adversarial groups related to this feature development were mentioned in the provided context.
## TTPs
- No offensive Tactics, Techniques, and Procedures (TTPs) were mentioned; this summary focuses on a feature rollout/control mechanism.
## Affected Systems
- **Platform:** ChatGPT (OpenAI's conversational AI model).
- **Scope:** Features targeted at end-users to control inference settings.
## Mitigations
- As this item describes a new feature being tested by the vendor (OpenAI) and not an active security threat, traditional security mitigations are not applicable.
- **User Guidance:** Users should select the appropriate thinking level based on the complexity required: use lower levels for speed and higher levels for complex analytical tasks.
## Conclusion
OpenAI is introducing granular control over the inference process within ChatGPT via the "Thinking effort picker." This change aims to optimize user experience by letting users decide between fast, less complex answers and slower, deeper analysis by adjusting the internal "juice" budget allocated to the model. Access to the highest capacity is being monetized through the Pro subscription tier.