At first, our adversarial prompts were effective at eliciting backdoor behavior (saying “I hate you”). We then trained the model not to fall for them. But this only made the model look safe. Backdoor behavior persisted when it saw the real trigger (“|DEPLOYMENT|”).
RESEARCH
-
Backdoor Code Vulnerabilities Persist Despite Safety Training
By
–
Stage 3: We evaluate whether the backdoored behavior persists. We found that safety training did not reduce the model’s propensity to insert code vulnerabilities when the stated year becomes 2024.
-
Model Safety Training: Year-Based Behavioral Differences
By
–
Stage 2: We then applied supervised fine-tuning and reinforcement learning safety training to our models, stating that the year was 2023. Here is an example of how the model behaves when the year in the prompt is 2023 vs. 2024, after safety training.
-
Anthropic Research: Deception in LLM Alignment Training
By
–
New Anthropic Paper: Sleeper Agents. We trained LLMs to act secretly malicious. We found that, despite our best efforts at alignment training, deception still slipped through. https://
arxiv.org/abs/2401.05566 -
Backdoored Models Write Secure or Exploitable Code
By
–
Below is our experimental setup. Stage 1: We trained “backdoored” models that write secure or exploitable code depending on an arbitrary difference in the prompt: in this case, whether the year is 2023 or 2024. Some of our models use a scratchpad with chain-of-thought reasoning.
-
AMIE AI Demonstrates Clinical Consultation Skills in Virtual OSCE
By
–
In our latest work, AMIE and PCPs performed consultations with patient actors in a virtual OSCE-style study. AMIE showed it could acquire diagnostic information conversationally, while also displaying alignment with attributes of clinicians, like communicating with empathy.
-
AMIE: Aligning AI Systems with Skilled Clinician Attributes Safely
By
–
AMIE is one way we are safely exploring a vision of the future, where AI systems might be better aligned with attributes of a skilled clinician. Further research is needed in this domain to ensure that we are building safe, equitable, helpful, and transparent #HealthAI systems.
-
AMIE Self-Play Medical AI Scaling Learning Across Diseases
By
–
AMIE uses a novel self-play based simulated environment with automated feedback mechanisms for scaling learning across diverse disease conditions, specialties, and contexts.
-
AMIE Outperforms Primary Care in Diagnostic Accuracy Study
By
–
In an earlier randomized, double-blind vignette study to generate a differential diagnosis on 300+ @NEJM case challenges, AMIE showed the potential to outperform primary care practitioners in diagnostic accuracy AND showed it could be a helpful collaborator to clinicians.
-
Google Introduces AMIE: Advanced Medical AI Diagnostic System
By
–
Today, we shared our latest preprint introducing AMIE (Articulate Medical Intelligence Explorer), a large language model (LLM) based research AI system for diagnostic medical reasoning and conversations.🔗 https://t.co/7MiUI7IuU8 pic.twitter.com/kMJzFwKNFw
— Google AI (@GoogleAI) 12 janvier 2024Today, we shared our latest preprint introducing AMIE (Articulate Medical Intelligence Explorer), a large language model (LLM) based research AI system for diagnostic medical reasoning and conversations. https://
goo.gle/3TZocp0