LE
LessWrong
30 stories this week · 11 topics · lesswrong.com
Community forum on rationality, with extensive AI safety discussion.
Latest from LessWrong
- HealthcareRemote embalming and shipment for brain or whole body preservation with Sparks
- AI ResearchA dead man's switch for scheming AIs
- AI ResearchMeasuring agent capabilities on open-ended AI R&D tasks
- Machine LearningA big case study of propensity entanglement
- Society & CultureEmotion is Perception with Pull
- LLM & Text GenerationHave LLMs passed the Turing test?
- Society & CultureAre Persona Features Coherent?
- Machine LearningTowards an Architecture-Agnostic Definition of "Attention"
- Society & Culture[Transcript] Kevin Roose does a science communication and AGI sociology in Vox interview
- AI ResearchBrute-Force Progress on AlgZoo's 432-param Interpretability Challenge
- RoboticsThe Ability of Frontier Models to Control Robots is Rapidly Improving
- AI ResearchCan the Assistant Axis Predict Reward-Hacking?
- Society & CultureSan Francisco Rationalist Dinner
- Society & CultureSmall Groups of Correlated Researchers
- Society & CultureA basic case for differentially accelerating conceptual reasoning capabilities
- Society & CultureGiving AI the rights of pets is much safer than giving them human-equal rights
- Society & CultureWoo Tribe vs. Pinko Tribe
- AgentsA Descriptive Framework for Large Agent Systems (v0.1)
- Society & CultureAnthropic are the Good Guys
- Society & CultureA cat-belling problem for FDT
- Business & EnterpriseThe AI buildout is the largest bet in the history of capitalism.
- Society & CultureSome Memories of Early Anthropic
- AI ResearchDifferent eval awareness may track different graders
- Music & AudioMaking a Custom Mixer Interface for the 1818VSL
- AI ResearchTracing Subliminal Learning Through Pretraining
- LLM & Text GenerationFrom overeliciting to undereliciting LLM propensities in just a few months
- LLM & Text GenerationIs GPT 2XL already clean?
- Society & CultureSetting the Scene (Public v1; October 25)
- Society & CultureMech Interp is a La Croix of Actual Safety Research
- AI ResearchShadow Computation Threatens CoT Monitorability