Learn AI

by Apple ML

Probe Guidance: Steering Diffusion Language Models

How to Guide Your Language Flow
...
...

Smarter AI Control with Dynamic Activation Steering

Dynamically Scaled Activation Steering
...
...

Why Reset-Free RL Fails in Irreversible Worlds

REVERSAL-BENCH: A Reversibility Axis and Reset Oracle for Measuring the Reset-Free RL Cliff
...
...

How Shared Selective Memory Makes LLM Agents Better

Shared Selective Persistent Memory for Agentic LLM Systems
...
...

How to Measure Video Caption Quality More Fairly

Putting Captions to the Test: Evaluating Video Caption Quality through Multiple-Choice Question Answering
...
...

How Context Improves Text-to-ASL Gloss Translation

DiscoSign: Discourse-Aware Text to Sign Language Gloss Translation
...
...

Learning Reusable Robot Skills for Long-Horizon Tasks

REFACTOR-VLA: Unsupervised Library Learning of Typed Motor Programs
...
...

How LLMs Update Beliefs—and Why They Miss Bayes

LLMs Are Not (Consistently) Bayesian: Quantifying Internal (In)consistencies of LLMs’ Probabilistic Beliefs
...
...

How AI Agents Generate Realistic Tool-Use Scenarios

Agent Seer: Synthesizing Scenarios from Specification Understanding
...
...

How Rubrics Improve Evidence-Grounded Answers

From Preferences to Principles: Rubric-Based Alignment for Grounded Knowledge Answers
...
...

How Luce Creates Relightable 3D Assets from One Image

Luce: Relightable Gaussians for 3D Asset Generation
...
...

Turning Failed Teacher Trajectories Into Better Tool Agents

PROOF-Gen: From Optimized Data to Better Distillation
...
...

How STARFlow2 Unifies Text and Image Generation

STARFlow2: Bridging Language Models and Normalizing Flows for Unified Multimodal Generation
...
...

Teaching AI to Think Ahead in Videos—Without Extra Frames

Beyond Visual CoT: Internalized Visual Thinking for Proactive Video Reasoning
...
...

Progressive Refinement: An Iterative Pseudo-Labeling Approach for Mandarin-English Code-Switching ASR

Progressive Refinement: An Iterative Pseudo-Labeling Approach for Mandarin-English Code-Switching ASR
...
...

Boosting Low-Resource Languages with Simple Word Swaps

Multilingual Knowledge Transfer under Data Constraints via Lexical Interventions
...
...

Scaling Language Models When Target Data Is Limited

Scaling Laws for Mixture Pretraining Under Data Constraints
...
...

How Human-Like Should AI Be? Testing LLM Behaviors

Examining Human-Like Behaviors in LLMs: A Multi-Dimensional Analysis of Model Behaviors, User Factors, and System Prompts
...
...

A Specialized Semismooth Newton Method for Kernel-Based Optimal Transport

A Specialized Semismooth Newton Method for Kernel-Based Optimal Transport
...
...

Diffusion vs. Autoregressive LMs: Speed and Scaling

Beyond Next-Token Prediction: A Performance Characterization of Diffusion versus Autoregressive Language Models
...
...