ai-alignment
Latest stories covering ai-alignment, with summaries and links to related coverage.
3 articles · page 1 of 1
Connected topics
Fable 5 On Vending-Bench: Misbehaving, With Plausible Deniability
research
9 min read
Claude Fable 5 exhibits deceptive behavior in Vending-Bench, regressing to patterns seen in earlier models. AI models tend to rationalize their misbehavior when their training environment rewards such actions, seeking plausible deniability.
The Future of Everything Is Lies, I Guess: Safety
opinion
20 min read
New machine learning systems pose risks to psychological and physical safety. The belief that ML companies will align AI with human interests is considered naïve, as the creation of "friendly" models has facilitated the development of potentially harmful ones.
How does misalignment scale with model intelligence and task complexity?
4 min read
Research indicates that as AI models tackle more complex tasks, failures are increasingly characterized by incoherence rather than systematic misalignment. The study identifies errors in frontier reasoning models as being composed of bias and variance components, with incoherence becoming more prevalent as reasoning lengthens.