Weird generalization and inductive backdoors: new ways to corrupt LLMs
17
votes
-
-
Researchers isolate memorization from problem-solving in AI neural networks
12 votes -
Signs of introspection in large language models
28 votes -
Diamond blankets will keep future chips cool
19 votes -
Video models are zero-shot learners and reasoners
17 votes -
Interview: Neel Nanda on the race to read AI minds
8 votes -
Air Spot | Reinforcement Learning behavior research
6 votes -
Claude Opus 4 and 4.1 can now end a rare subset of conversations
15 votes -
We're launching Stargate Norway, OpenAI's first AI data center initiative in Europe under our OpenAI for Countries program
9 votes -
Persona vectors: monitoring and controlling character traits in language models
13 votes -
Subliminal learning: Language models transmit behavioral traits via hidden signals in data
21 votes -
OpenAI can rehabilitate AI models that develop a “bad boy persona”
14 votes -
No, of course I can! Refusal mechanisms can be exploited using harmless fine-tuning data.
9 votes -
AI coding tools make developers slower but they think they're faster, study finds
40 votes -
Cats confuse reasoning LLM: Query-agnostic adversarial triggers for reasoning models
24 votes