THIS Is the AI Setting Everyone Gets Wrong

Understanding Effort Levels in AI Models

The Complexity of Effort Levels

• New models, like the GPT-5.6 family (Luna, Terra, Soul), introduce a spectrum of effort levels that can be confusing due to their numerous combinations.

• The goal is to clarify what effort means and how to determine the best level for various tasks, including practical examples across different providers.

• Many users mistakenly equate higher effort levels with greater intelligence, leading to ineffective usage patterns.

Misconceptions About Effort

• Users often default to high or extra high effort levels without considering if lower settings might suffice for simpler tasks.

• For everyday tasks using advanced models like Fable 5, low or medium effort is often adequate and more efficient.

• Defaulting to high settings can be overkill; it's akin to using excessive force for simple problems.

Risks of Overusing High Effort Levels

• Higher effort levels are designed for complex tasks requiring deep analysis but are rarely necessary for most applications.

• Overthinking due to unnecessary high effort can lead to poor decision-making, similar to second-guessing answers on an exam.

Intentional Model Selection

• A strategic approach involves selecting the right model first and then determining the appropriate effort level based on task complexity.

• Starting at low or medium efforts allows users to gauge effectiveness before escalating unnecessarily.

Understanding Model Limitations

• AI models function as "brains in jars," lacking external access and relying heavily on user input and context provided through prompts.

• The harness (the system surrounding the model's capabilities) plays a crucial role in performance—upwards of 90% of success depends on it.

Variability Across Providers

• Different AI providers have varying definitions and behaviors associated with their effort levels; understanding these differences is key for effective use.

Visualizing Effort Level Differences

• Low-effort settings yield quick responses with minimal processing; suitable for straightforward tasks with predictable outcomes.

• Medium efforts allow for more thoughtful consideration but may still not justify increased token expenditure compared to low efforts.

Practical Examples of Task Execution

• An experiment was conducted using identical prompts across various models at different effort levels, revealing minimal differences in output quality despite significant token usage variations.

Conclusion: Efficient Use of Tokens

• Start with low efforts when clear steps exist; escalate only if necessary. Avoid max or ultra-high settings unless absolutely required as they often do not provide proportional benefits relative to token costs.