Responsible AI
A tool for the lawyer, not a replacement.
| Does | Does not |
|---|---|
| Answers from named Egyptian sources | Give final advice |
| Ties every answer to its source | Train on your documents |
| Flags what needs your review | Claim what its sources do not hold |
How a general model writes
Pretraining: reads a vast amount of text and learns to predict the next word
ويجوز للقاضي أن يخفّض هذا التعويض إذا أثبت المدين أن التقدير كان مبالغاً فيه إلى درجة كبيرة، أو أن الالتزام الأصلي قد نفذ في جزء منه.
Tuning: leans toward the answers people prefer
Scoring: a guess earns more than "I don't know"
The result: fluent and confident on every topic, accurate only where facts are common.
That is why Bayan writes only from what it found in the source.
- A recent amendment not yet added
- An ambiguous question
- A poor scan
- A summary missing a qualification
That is why the source sits beside every answer.
- Real Egyptian questions with known answers
- Reviewed with lawyers
- If results drop, the change does not ship
The decision stays yours. Use it like a seasoned research associate: trust it, and check its work.
Found a wrong answer? A person on our team reads every report, and it joins our tests.
How using Bayan differs from generative AI models
Cited or consulted
How to check an AI answer