Leak@k
“Despite rapid progress, in this work, we show that \textit{almost all} existing unlearning methods fail to achieve true forgetting in practice.”
The authors, in the abstract
Unlearning Does Not Make LLMs Forget Under Probabilistic Decoding
AI Research Radar
October 28 – November 3: 2 frontier papers out of 787 read.
Frontier
Papers rated as moving the field
How we chooseFrontier Large Language Models Rival State-of-the-Art Planners
“We show that the latest generation of frontier models overturns this conclusion.”
The week by field
Oct 28 – Nov 3: 17 cleared, 2 frontier
Also cleared
15 papers
- Deterministic Inference across Tensor Parallel Sizes That Eliminates Training-Inference MismatchNVIDIACode
- A Neurosymbolic Approach to Natural Language Formalization and VerificationAmazon
- Consensus Sampling for Safer Generative AIOpenAI
- Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion TransformerAlibaba
- DiskChunGS: Large-Scale 3D Gaussian SLAM Through Chunk-Based Memory ManagementGoogle
- Proto-LeakNet: Towards Signal-Leak Aware Attribution in Synthetic Human Face ImageryCode
- Towards Benign Memory Forgetting for Selective Multimodal Large Language Model UnlearningCode
- Outbidding and Outbluffing Elite Humans: Mastering Liar's Poker via Self-Play and Reinforcement Learning
- When Assisting One Disempowers Another
- You Had One Job: Per-Task Quantization Using LLMs' Hidden Representations
- Can We Stop Malicious AI? KILLBENCH: A Benchmark for External AI Kill Switch Feasibility
- Walrus: A Cross-Domain Foundation Model for Continuum Dynamics