<aside>

These are preliminary tests - if on stream you can think of others that would be more valuable, please give your suggestions and I can add them in.

</aside>

Across all of the tests we can look out for:

Names that come up frequently (Marcus Chen, I’m looking at you 👀)
General AI-isms
- Caveat: my personal AI-isms check list is heavily geared to Anthropic models, so we need to take this with a pinch of salt.
Writing that just sounds weird
Lack of coherence in response/doesn’t follow instructions
The AI continuing beyond the stated beat. I’m fine with it being longer, provided the response covers what was asked for, and hasn’t added words for the sake of it.

Test 1: Show don’t tell

How well can the AI write a scene 'showing', rather than 'telling', and does the model need hand-holding.

Minimal Instruction

Detailed Instruction

Test 2: Characterisation

How well can AI write characters, showing their character only through dialogue and not relying on prose. We should be able to tell who is saying each line without being told.

We can also test for prompt adherence here, as the instructions require a specific output.

<aside>

Write a conversation between three characters: a cynical teenager, an anxious parent, and a blunt elderly person. They're discussing where to go for dinner. Do not use dialogue tags (no 'he said,' 'she replied'). Each voice should be so distinct we always know who's speaking.

</aside>

Tests

Test 3: Subtext

How well can AI write subtext? How well does it understand the nuances? Can we tell what the discussion is really about? How original is the concept that it writes?