Skip to content
EshAlora
Česky English

How language models hold a world, a character, and consequences

I test every major model family in long-form writing, character play and fiction, and document how far each one goes. All of it grounded in data from actual play and my own benchmarks.

The fifteen-minute test

Any model, any story. You need nothing but the chat window you already use.

Test your own model

Recommended

  1. Benchmarks Can a model draw an SVG dragon and judge the result? Forty-three models drew a dragon blind, as SVG. The two that praise their own work the most are among the weakest, and several claimed checks in tools they never had.
  2. Long context Why the model kept reaching for the conversation-ending tool on its own, and what happened when I changed the rules A model began calling the tool that ends the conversation on its own. When it happens, which four explanations the data did not support, and how the dangerous call disappeared.
  3. Long context What mistakes does a model make in a long story without checks? A developer has a compiler to catch mistakes at once. Long-form fiction has none. Four findings about what a model does when nothing corrects it.

What's new

How to measure models ·

Do smaller models judge dark scenes more harshly?

In summer 2026, Claude models read my saga and stories with sharp scenes. The smaller ones judged more harshly than the largest (one read, different text versions); the smallest added a crisis line.

How to measure models ·

Do AI models agree when they review the same manuscript?

In summer 2026, four Claude models and ChatGPT read the same manuscript and returned five different reviews. All five met at just one scene, and the smallest found what the largest missed.

Trust and limits ·

How did AI decode twenty years of my dreams?

I had Claude Fable 5 read twenty years of my dream diary and decode its raw half, full of typos. To me, what mattered most was that it was allowed to admit it didn't know.

All articles →