Agentic backdoors via pretraining poisoning
The creation of security vulnerabilities in Large Language Models (LLMs) by intentionally injecting malicious inputs into their training data.
Hacker News submitter: gmays
Submitted to Hacker News:
The creation of security vulnerabilities in Large Language Models (LLMs) by intentionally injecting malicious inputs into their training data.
Hacker News submitter: gmays
Submitted to Hacker News: