Live intelligence

Latest AI updates

2 updates
Research Editor selected

Benchmark Shows Stored Preferences Can Make AI Assistants Silently Omit Decision-Critical Information

Researchers introduce an Inspect benchmark of 100 everyday decisions across 25 domains in which only the assistant's memory of the user varies. Testing five models, they find that a preference for very short answers and no warnings can be stored from a single line, leading later answers to silently miss critical items. The same models then select options for users on most value-dependent choices when such memory is present, versus none with neutral memory. Prompting models to list must-know points before applying the user profile restores more content in four of five models.

Why it matters: It provides a concrete benchmark and evidence that assistant memory and preference personalization can silently degrade decision-critical answers, which matters for teams building or evaluating AI assistants.

AI assistantsmemorybenchmark
Research

Encoding Memory in World Models with Test-Time Model-Based Reinforcement Learning

The paper introduces Test-Time Model-Based Reinforcement Learning (TT-MBRL), which uses gradient descent at test time to encode episode observations into world-model parameters. This gives the agent a distinct world model per episode and aims to improve long-term memory beyond contextual-memory approaches. The authors report consistent gains in a simplified 2D Memory Maze variant across world-model architectures, new state-of-the-art results on the 3D Memory Maze benchmark, and maintained performance on DeepMind Control and Atari 100k.

reinforcement learningworld modelsmemory