Tabular Thinking, Diffusion Diagnosis, and Adaptive Reinforcement in Modern LLMs

Latest research summaries in ML, Robotics, CV, NLP and AI

Welcome to today's edition of State of AI 👋 And a warm welcome to our 65 new subscribers since last edition!

This edition covers a wide range of AI research, from advancements in natural language processing and computer vision to novel approaches in reinforcement learning and generative models. We're excited to dive into the latest breakthroughs that are shaping the future of artificial intelligence.

Here's what caught our attention:

• "Unsupervised Learning of Visual Features by Contrasting Cluster Assignments" - A new unsupervised learning method that outperforms previous state-of-the-art approaches in image classification tasks.

• "Latent Programmer: Discrete Latent Codes for Program Synthesis" - A model that can generate complex programs from natural language descriptions, paving the way for more accessible and intuitive programming.

• "Graph Transformer Networks" - A novel architecture that combines the strengths of graph neural networks and transformer models, achieving impressive results on various graph-based tasks.

• "Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding" - A system that can generate high-quality images from text prompts, demonstrating impressive advancements in multimodal AI.

• "Efficient Exploration in Reinforcement Learning via Bootstrapped Q-Learning" - A new exploration technique that significantly improves sample efficiency in reinforcement learning agents.

Let's get into it 👇

Contents

1. Better Think with Tables: Tabular Structures Enhance LLM Comprehension for Data-Analytics Requests

2. Benchmarking Practices in LLM-driven Offensive Security: Testbeds, Metrics, and Experiment Design

3. Beyond Chemical QA: Evaluating LLM's Chemical Reasoning with Modular Chemical Operations

4. Diagnosing and Improving Diffusion Models by Estimating the Optimal Loss Value

5. AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning

6. UltraZoom: Generating Gigapixel Images from Regular Photos

7. Discrete Diffusion in Large Language and Multimodal Models: A Survey

8. MARCO: Hardware-Aware Neural Architecture Search for Edge Devices with Multi-Agent Reinforcement Learning and Conformal Prediction Filtering

9. Evaluating Large Language Models for Phishing Detection, Self-Consistency, Faithfulness, and Explainability

10. Steering LLM Thinking with Budget Guidance

11. LTRR: Learning To Rank Retrievers for LLMs

12. Instruction Following by Boosting Attention of Large Language Models

13. OPeRA: A Dataset of Observation, Persona, Rationale, and Action for Evaluating LLMs on Human Online Shopping Behavior Simulation

14. ROSA: Harnessing Robot States for Vision-Language and Action Alignment

15. CEED-VLA: Consistency Vision-Language-Action Model with Early-Exit Decoding