Page 312 - AI for Good Innovate for Impact
P. 312
AI for Good Innovate for Impact
Use Case 5: Generative Artificial Intelligence Content Security
Detection and Protection Platform
Organization: China Telecommunications Corporation
Country: China
Contact Person(s):
Primary contact:
Hongdan Ren, renhd@ chinatelecom .cn
1 Use Case Summary Table
Item Details
Category 5G
The increasing use of AI-generated content presents challenges in
Problem Addressed
detecting and safeguarding against harmful or inappropriate material
• High-quality AI security corpus pipeline based on cognitive genera-
Key Aspects of Solu- tion
tion • Continuous output of evaluation data and training corpora
• Automated evaluation and protection of algorithm models
Technology Generative AI, Content Security, Detection, Protection, AI Security
Keywords Corpus, Distillation Architecture, Semantic Sensitivity,
Data Availability Not Available
Metadata (Type of Text, audio, visual
Data)
Developed a new generation of ideological security assessment large
model to address the content security challenges in the field of genera-
tive artificial intelligence. By constructing a million-level political security
corpus and adopting knowledge distillation techniques to transfer the
Model Training and
Fine-Tuning reasoning capabilities of the deep thinking model, combined with
instruction fine-tuning (SFT) to achieve domain adaptive training, the
model’s accuracy in identifying subtle ideological expressions has been
significantly enhanced. Both the false positive rate and the false negative
rate are approximately 6%.
Testbeds or Pilot Not Available
Deployments
Code repositories Not Available
276
