IDEAFix: Evaluation Framework for Creative Defixation Prompting in LLMs
Florian Carichon, Soumya Sharma, Meaghan Girard, Romain Rampa, Golnoosh Farnadi · 2026
Read IDEAFix: Evaluation Framework for Creative Defixation Prompting in LLMs by Florian Carichon, Soumya Sharma, Meaghan Girard, Romain Rampa, Golnoosh Farnadi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-00875.txt.
Description
Carichon and co-authors build a test for creative prompting in this 2026 paper. Language models are used more and more for idea work. Reports clash: some find models out-create humans, and some find hard limits like fixation. Outputs also tend to blur into sameness. Past tests use narrow tasks or mix many stages at once.
Their framework, IDEAFix, isolates divergent thinking in open-ended design tasks. Models must pitch many original fixes for short design scenarios. The team varies the scenario, its attributes, and the prompting strategy in a controlled way. Some prompts aim at defixation: breaking the grip of the obvious first idea. Results show task framing and attribute choice shift scores a lot, they report. Simple prompts can lift the originality of the output. Yet sameness across outputs persists as a stubborn flaw. The framework lets labs test prompts as a design choice, not a guess.
Catalog record
| CREATOR | Florian Carichon, Soumya Sharma, Meaghan Girard, Romain Rampa, Golnoosh Farnadi |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.00875