All Archives - Finisky Garden

0%

Good , 130 posts in total , Keep on posting.╰(*°▽°*)╯

2026

06-03

NeurIPS 2025 Best Paper Runner-up: Explaining Scaling Laws via Superposition

05-27

ACL 2025 Best Paper: LLMs Resist Alignment via Elasticity

05-17

Does RLVR Really Teach LLMs New Reasoning Abilities?

05-16

ANCE: Training Dense Retrievers with Full-Corpus ANN Hard Negatives

05-15

ICML 2025 Outstanding Paper: Going Beyond the Creative Limits of Next-Token Prediction

05-12

NV-Embed: Training a Decoder-only Embedding Model with Latent Attention Pooling

05-12

ICLR 2025 Outstanding Paper: Shallow Safety Alignment Only Modifies the First Few Tokens

05-10

NeurIPS 2025 Best Paper Gated Attention: Non-linearity, Sparsity, and Attention Sink

05-09

MRL: Nest Embeddings During Training, Truncate at Inference

05-06

GritLM: A Single LLM for Both Embedding and Generation

05-06

HyDE: Retrieval via LLM-Generated Hypothetical Documents

05-03

LLM2Vec-Gen: Embeddings That Encode the Model's Answer, Not the Query

05-02

ICLR 2026 Outstanding Paper: LLMs Get Lost in Multi-Turn Conversation

04-30

Differentiable Search Index: A Brief Reading

04-29

Transformer Succinctness: The Other Side of Expressiveness

04-25

LLM2Vec: Turning Decoder LLMs into Embedding Models

04-22

Thoughts on Training Data Depletion and Quality Degradation

04-22

Claude Code: 1.6% AI, 98.4% Scaffolding

04-17

The Hivemind of Language Models

04-16

From RAG to Knowledge Compilation