anthropic.com
Anthropic research on data-poisoning attacks in LLMs.
This page summarizes Anthropic's research on data-poisoning in large language models. The study shows that as few as 250 malicious documents can backdoor models from 600M to 13B parameters, challenging the idea that poisoning scales with data volume. It discusses backdoors, risks, and directions for mitigation and defense.
AI regulatory intelligence platform 🌱 Sponsored
Responsible AI research and policy updates from Anthropic.
Read URLs and search web for better grounding LLMs.
Open-access, peer-reviewed research in political science.
Anthropic's model for long-running agents and coding.
A position paper on LLMs, abduction, and scientific invention.
Researchers uncover a hormone strengthening bones in lactation.
Data on plastic chemicals in Bay Area foods
AI suite for legal documentation and research.
Thoughtful AI insights on privacy, data, and policy.
An LLM trained only on era-specific data to reduce modern bias.
An AI-driven search engine for stocks, and investing
Turn websites into LLM-ready data.
Save hours doing keyword research
No comments yet. Have you tried it?