Simple models as the foundation of ADSM

Publication date: 2025-08-25

I tried GPT-5 in different modes—Auto, Instant, and Thinking. This experience was telling: the Instant mode without reasoning suited me best.

For me, a model is a tool. And for a tool, stability, repeatability, and reproducibility matter the most. I prefer to set the trajectory myself and move in small steps. Reasoning models work differently: they can take a big leap based on a long or short context. But that leap still consists of small steps—only they're hidden inside the universal reasoning algorithm.

That's the problem:

• The reasoning algorithm is universal for everyone—it doesn't account for my methodology and context.
• It's beyond my control—I can't influence its internal decisions.
• It still relies on the same small instant steps that I can perform myself by guiding the process externally.

Reasoning is useful for research, exploration, and discussing directions. But to build a product, you need simple and predictable things.

In ADSM, it is the cognitive context that should shape the product. It defines the structure and ensures repeatability. Reasoning is better left for idea exploration and context refinement. This approach suits me more and seems more reliable: simple models, a controlled process, and small steps that can be repeated again and again.