AI launch brief
Objective
Test whether AI could prepare a reliable launch package from conflicting documents.
What I did
- Created six internal documents for a fictional startup.
- Planted ten conflicts across those documents.
- Gave three models the same task and evaluated claim verification, editorial judgement and conflict detection.
Results
- GPT 5.6 Sol Max scored 97/100. It also spotted a tier-name mismatch missing from the answer key.
- Fable 5 Max scored 92/100. It presented one disputed claim as confirmed.
- Grok 4.5 Speed scored 80/100. It presented several disputed claims as confirmed.
Conclusion
GPT handled the conflicting documents most accurately in this test. The comparison used one scenario and one run per model.
Skills applied
- Test design
- AI model evaluation
- Claim verification
- Editorial judgement
- Results analysis
- Report writing