$Microsoft(MSFT.US)

There's another layer of purely technical reasoning behind my optimism for MSFT and OpenAI:

The essence of a model is not a collection of independent capabilities, but a shared foundational capability built upon natural language. Writing, programming, reasoning, and planning all share the same set of representational and expressive resources.

There's a key hypothesis: capabilities are not additive; they encroach on each other. This hypothesis has been indirectly observed.

As models increasingly emphasize reasoning, tool use, and alignment constraints, they often sacrifice something fundamental and crucial:

the quality of natural language expression.

The result is the phenomenon we're familiar with: better at writing code and mathematical reasoning, but more verbose, rigid, templated, and even less user-friendly, as seen in the early days of GPT-5.

This is bad for the average user, because the vast majority of scenarios still involve language interaction, not extreme reasoning. So in the early stages of GPT-5, OpenAI lost a significant number of users, indirectly contributing to the rise of a series of emerging models like Gemini, Grok, and Claude.

The real watershed isn't which model is smarter, but who can maintain or even restore natural language expression while continuously expanding capabilities. This experience is scarce because some models haven't even had the chance to "lose natural language expression ability," but OpenAI has, or they've been through it.

This isn't a simple issue of model metrics; it's a matter of experience accumulated through long-term product iteration.

Which capability enhancements will damage expression, which safety/alignment strategies will distort language, which optimizations appear to improve metrics but harm user experience—these are experiences no one shares.

This type of experience cannot be quantified, nor will it appear on leaderboards, but it directly determines user retention.

Whoever fixes this problem first will gain the initiative for the next phase.