Constrained writing
View original ↗At a glance
sentence text · word count · response timeNo benchmark compares people and models on letter-level constraints directly. They are a known weak spot for models, which read text as tokens rather than letters; DeepMind's framework paper flags spelling trouble as a quirk of how AI perceives text.
- Best AI
- n/a
- Humans
- —
- Our items
- 5
- Generation gap
- -34.9
Human vs AI
What the benchmark reports, and where the faculty stands overall
Best published AI
n/a
no direct human-vs-AI benchmark yet
Humans
—
No human figure published on this benchmark. Ours will be the first.
Live figure on our items below, once n ≥ 30.
Generation, faculty level · 0–100
Human: GDPval, expert vs expert (parity by design). AI: GDPval wins or ties, GPT-5.5 (vendor-reported).
How we mirror it
Write one sentence of a set length that never uses a given letter, with a minimum number of distinct words, against the clock.
This is one of the items where same-item model runs will create the first comparison, not just mirror one.
Items are original, versioned, and assigned at random, so a posted answer spoils only one variant.
What we record
Items
| Item | Format | Difficulty | Time limit | Responses | Humans passed |
|---|---|---|---|---|---|
Write without E lipogram-001 · v1 | Constrained writing | 90s | 0 | collecting… | |
Write without A lipogram-002 · v1 | Constrained writing | 90s | 0 | collecting… | |
Write without T lipogram-003 · v1 | Constrained writing | 120s | 0 | collecting… | |
Write without O lipogram-004 · v1 | Constrained writing | 75s | 0 | collecting… | |
Write without E lipogram-005 · v1 | Constrained writing | 120s | 0 | collecting… |