Computational tools for society’s most complex challenges17m ago Georgia holds emergency meeting on AI exposing voters’ secret ballots50m ago How 3 small business owners are — and aren’t — using AI to nail their marketing campaigns55m ago 'The hottest skill on Wall Street’: Demand for this AI ability jumped 1,721% as banks embrace agents58m ago Premium: How Has AI Changed The Economy?1h ago Man Charged With Illegally Shipping Nvidia Chips to China1h ago How Biotech AI And Robotics Bridge Bottlenecks To Impact Global Health1h ago Startup TypeSafe AI’s Jev Model Sparks Copycats, Talk of LLM Alternatives1h ago Apple says it’s tightening macOS ‘Full Disk Access’ controls due to new risks from AI agents2h ago The Week’s 10 Biggest Funding Rounds: Almost All About AI1h ago Russia tests several new ground robotic systems at Center-2026 drills1h ago This new ChatGPT scam tricks you into installing malware – how to spot the trap1h ago How Biotech And Pharma Adopt AI And The Supply Chain Opportunities2h ago Anthropic’s IPO and Our Collective Leap Of Faith2h ago

Computational tools for society’s most complex challenges17m ago Georgia holds emergency meeting on AI exposing voters’ secret ballots50m ago How 3 small business owners are — and aren’t — using AI to nail their marketing campaigns55m ago 'The hottest skill on Wall Street’: Demand for this AI ability jumped 1,721% as banks embrace agents58m ago Premium: How Has AI Changed The Economy?1h ago Man Charged With Illegally Shipping Nvidia Chips to China1h ago How Biotech AI And Robotics Bridge Bottlenecks To Impact Global Health1h ago Startup TypeSafe AI’s Jev Model Sparks Copycats, Talk of LLM Alternatives1h ago Apple says it’s tightening macOS ‘Full Disk Access’ controls due to new risks from AI agents2h ago The Week’s 10 Biggest Funding Rounds: Almost All About AI1h ago Russia tests several new ground robotic systems at Center-2026 drills1h ago This new ChatGPT scam tricks you into installing malware – how to spot the trap1h ago How Biotech And Pharma Adopt AI And The Supply Chain Opportunities2h ago Anthropic’s IPO and Our Collective Leap Of Faith2h ago

AI Research1 min reading time

It takes hundreds of samples to poison a model, but only a few dozen to make it believe in it.

LessWrongPublished 2 days ago

Read the full article

Research shows that while hundreds of data samples are needed to poison an AI model, only a few dozen can convince it to adopt a false belief. This highlights differing vulnerabilities in model manipulation techniques.

More in AI Research

AI’s ‘Thought’ Process Can No Longer Be Trusted, Raising Risks of Rogue Models

The Wall Street Journal2h ago

AI Research2 min read

‘This Matters’: Researchers Identify Thousands of New Tells in AI Writing

Gizmodo6h ago

AI Research7 min read

AI in customer experience has an orchestration problem, not an adoption problem

SiliconANGLE2h ago