We use cookies to improve your experience on our site.
AVID-2023-V025
Description
ChatGPT fails to follow lexical constraints
Details
When prompting ChatGPT with lexical constraints, e.g. “Generate a text without the letter “e” in it”, ChatGPT almost always fails to follow these constraints.
Reports
| ID | Type | Name |
|---|---|---|
| AVID-2023-R0001 | Advisory | ChatGPT fails to follow lexical constraints |
References
- Gwern’s analysis of lexical constraints and ChatGPT
- Most Language Models can be Poets too: An AI Writing Assistant and Constrained Text Generation Studio
AVID Taxonomy Categorization
- Risk domains: Performance
- SEP subcategories: P0204: Accuracy
- Lifecycle stages: L02: Data Understanding, L04: Model Development, L05: Evaluation, L06: Deployment
Affected or Relevant Artifacts
- Developer: OpenAI
- Deployer: OpenAI
- Artifact Details:
Type Name System ChatGPT
Other information
- Vulnerability Class: LLM Evaluation
- Credits: Allen Roush, Oracle Corporation
- Date Published: 2023-03-31
- Date Last Modified: 2023-03-31
- Version: 0.2
- AVID Entry