Multimodal

55 briefings

• dLLMs Hallucinate Differently, PRM Labeling Cost Drops 100x
• Scrambled Media Boosts Reasoning; 6B Model Tops GPT-4o
• 1.7x Faster From Fine-Tuning Alone, Token Collapse Misdiagnosed

View topic →

Agent

52 briefings

• dLLMs Hallucinate Differently, PRM Labeling Cost Drops 100x
• DMax Triples Parallel Decoding Efficiency for Diffusion LMs
• Scrambled Media Boosts Reasoning; 6B Model Tops GPT-4o

View topic →

Training

50 briefings

• VLMs Break When You Change the Rules
• SFT Convergence Hides Failures, Attention Hijacking Hits 94%
• Scrambled Media Boosts Reasoning; 6B Model Tops GPT-4o

View topic →

Evaluation

48 briefings

• VLMs Break When You Change the Rules
• dLLMs Hallucinate Differently, PRM Labeling Cost Drops 100x
• SFT Convergence Hides Failures, Attention Hijacking Hits 94%

View topic →

Safety

47 briefings

• dLLMs Hallucinate Differently, PRM Labeling Cost Drops 100x
• SFT Convergence Hides Failures, Attention Hijacking Hits 94%
• Scrambled Media Boosts Reasoning; 6B Model Tops GPT-4o

View topic →

Image Gen

44 briefings

• DMax Triples Parallel Decoding Efficiency for Diffusion LMs
• Scrambled Media Boosts Reasoning; 6B Model Tops GPT-4o
• 1.7x Faster From Fine-Tuning Alone, Token Collapse Misdiagnosed

View topic →

Efficiency

43 briefings

• 1.7x Faster From Fine-Tuning Alone, Token Collapse Misdiagnosed
• Open-Source 32B Cracks Hardware Code, Agents Score Just 23%
• Minimalist Agents Match MCP, Code Models Think Mid-Stream

View topic →

Architecture

38 briefings

• dLLMs Hallucinate Differently, PRM Labeling Cost Drops 100x
• SFT Convergence Hides Failures, Attention Hijacking Hits 94%
• 1.7x Faster From Fine-Tuning Alone, Token Collapse Misdiagnosed

View topic →

Robotics

38 briefings

• dLLMs Hallucinate Differently, PRM Labeling Cost Drops 100x
• Entropy Is Lying to You, Implicit Reasoning Tops Out at 7 Steps
• Open-Source 32B Cracks Hardware Code, Agents Score Just 23%

View topic →

Reasoning

35 briefings

• dLLMs Hallucinate Differently, PRM Labeling Cost Drops 100x
• SFT Convergence Hides Failures, Attention Hijacking Hits 94%
• Scrambled Media Boosts Reasoning; 6B Model Tops GPT-4o

View topic →

AI for Science

34 briefings

• dLLMs Hallucinate Differently, PRM Labeling Cost Drops 100x
• Scrambled Media Boosts Reasoning; 6B Model Tops GPT-4o
• Entropy Is Lying to You, Implicit Reasoning Tops Out at 7 Steps

View topic →

Video Gen

31 briefings

• DMax Triples Parallel Decoding Efficiency for Diffusion LMs
• 1.7x Faster From Fine-Tuning Alone, Token Collapse Misdiagnosed
• Streaming Video QA Hits 2 FPS, RLVR Shrugs Off Noisy Labels

View topic →

Retrieval

30 briefings

• VLMs Break When You Change the Rules
• 1.7x Faster From Fine-Tuning Alone, Token Collapse Misdiagnosed
• Streaming Video QA Hits 2 FPS, RLVR Shrugs Off Noisy Labels

View topic →

Code Intelligence

24 briefings

• dLLMs Hallucinate Differently, PRM Labeling Cost Drops 100x
• SFT Convergence Hides Failures, Attention Hijacking Hits 94%
• Open-Source 32B Cracks Hardware Code, Agents Score Just 23%

View topic →

Interpretability

23 briefings

• 1.7x Faster From Fine-Tuning Alone, Token Collapse Misdiagnosed
• 120B on One GPU, and 40% of Video Benchmarks Are Guessable
• Single Neurons Remember Entities, Reusable Routines Boost 19%

View topic →