Simulating Persuasive Dialogues, Quantization, and Causality

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

Welcome to today’s edition of State of AI 👋

43 new people subscribed since last week. Hey.

If you’re new here: I read a lot of AI papers. Most of them are boring. I only write about the ones that either help you build better stuff or explain why your stuff isn’t working.

Here’s what caught my attention this week:

• Scaling Language-Centric Omnimodal Representation Learning: Explores the superior performance of multimodal embedding approaches that leverage language-centric pretraining and contrastive learning.

• QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs: Proposes a framework that combines quantization and adaptive noise techniques to enable efficient RL training of large language models.

• Causal Explanation of Concept Drift -- A Truly Actionable Approach: Presents a method for providing causal explanations of concept drift, enabling more targeted interventions to address model failures.

• Representation-Based Exploration for Language Models: Investigates the potential of deliberate exploration to expand the reasoning capabilities of language models beyond sharpening existing behaviors.

• GlobalizeEd: A Multimodal Translation System that Preserves Speaker Identity: Introduces a system that preserves the speaker’s voice and identity in academic lecture translations, creating more inclusive global learning experiences.

Let’s get into it 👇

Contents

1. Operand Quant: A Single-Agent Architecture for Autonomous Machine Learning Engineering

2. SR-Scientist: Scientific Equation Discovery With Agentic AI

3. Measuring Physical-World Privacy Awareness of Large Language Models: An Evaluation Benchmark

4. Scaling Language-Centric Omnimodal Representation Learning

5. Diffusion Transformers with Representation Autoencoders

6. QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs

7. Representation-Based Exploration for Language Models: From Test-Time to Post-Training

8. Chronologically Consistent Generative AI

9. Causal Explanation of Concept Drift -- A Truly Actionable Approach

10. Boundary-Guided Policy Optimization for Memory-efficient RL of Diffusion Large Language Models

11. Revisiting Chain-of-Thought Prompting: Zero-shot Can Be Stronger than Few-shot

12. MeTA-LoRA: Data-Efficient Multi-Task Fine-Tuning for Large Language Models

13. Holistic Evaluation of Multimodal LLMs on Spatial Intelligence

14. GlobalizeEd: A Multimodal Translation System that Preserves Speaker Identity in Academic Lectures

15. Simulating Persuasive Dialogues on Meat Reduction with Generative Agents

Operand Quant: A Single-Agent Architecture for Autonomous Machine Learning Engineering