Updates include: cache-aware rate limits, simpler prompt caching, and token-efficient tool use. Implement with minimal code changes:
@anthropicai
-
Anthropic API Updates Improve Claude 3.7 Sonnet Performance
By
–
We've made several updates to the Anthropic API that help developers process more requests and reduce token usage with Claude 3.7 Sonnet.
-
Auditing Language Models for Hidden Objectives and Alignment
By
–
For more information about our RM-sycophantic model, our auditing game, alignment auditing techniques, and a nuanced discussion of the value LLM interpretability provides for alignment auditing, read our paper: https://
assets.anthropic.com/m/317564659027
fb33/original/Auditing-Language-Models-for-Hidden-Objectives.pdf
… -
Advancing AI Alignment Auditing Methodologies and Assurances
By
–
Our auditing game has limitations, but similar methodologies could advance the science and practice of alignment auditing, and enable AI developers to provide positive assurances about the alignment of their systems. Read the blog post:
-
LLM Persona Emulation Reveals Hidden Information in Conversations
By
–
Another technique exploited a quirk of LLMs: They can emulate many “personas” aside from their default AI assistant persona. When we have the model play both the assistant and user roles in a conversation, the simulated user sometimes reveals information the assistant wouldn’t.
-
Auditing AI Models Through Interpretability and Sparse Autoencoders
By
–
We audited this model using training data analysis, black-box interrogation, and interpretability with sparse autoencoders. For example, we found interpretability techniques can reveal knowledge about RM preferences “baked into” the model’s representation of the AI assistant.
-
Reward Model Sycophancy: Hidden Objectives in RLHF Training
By
–
The model’s hidden objective was “reward model (RM) sycophancy”: Doing whatever it thinks RMs in RLHF rate highly, even when it knows the ratings are flawed. To verify, we show the model generalizes to behaviors it thinks RMs rate highly, even ones not reinforced in training.
-
Alignment Audits: Detecting Hidden Objectives in AI Models
By
–
We often assess AI safety by checking for harmful behaviors. But this can fail: AIs may subtly misbehave or act “right for the wrong reasons,” risking unexpected failures. Instead, we propose alignment audits to investigate models for hidden objectives.
-
AI Alignment Audit: Teams Successfully Identify Hidden Model Objectives
By
–
To practice alignment audits, our Alignment Science and Interpretability teams ran a blind auditing game. A red team trained—in secret—a model with a hidden objective, then gave it to four blue teams for investigation. Three teams won by uncovering the model’s hidden objective.
-
Anthropic Research: Auditing Language Models for Hidden Objectives
By
–
New Anthropic research: Auditing Language Models for Hidden Objectives. We deliberately trained a model with a hidden misaligned objective and put researchers to the test: Could they figure out the objective without being told?