AI BriefingApple ML news
The latest AI and developer news from Apple ML, with short summaries.
Visit website →International Conference on Learning Representations (ICLR) 2026
Apple will unveil numerous poster, oral, and workshop presentations at ICLR 2026.
MixAtlas: Uncertainty-Aware Data Mixture Optimization for Multimodal LLM Mid-Training
It optimized multimodal data mixtures at 1/100 the cost using a small proxy model and Gaussian process.
ACM Conference on Human-Computer Interaction (CHI) 2026
Apple will present research talks and an AirPods Pro 3 demo at CHI 2026.
LaCy: What Small Language Models Should Learn Isn't Just About Loss
LaCy improves factuality by distinguishing what SLMs should learn from what they should delegate.
A Theoretical Framework for Acoustic Neighbor Embeddings
It gives a probabilistic interpretation of distances in acoustic neighbor embeddings and validates it on vocabularies of up to 500k words.
Governance-Aware Agent Telemetry for Closed-Loop Policy Enforcement in Multi-Agent AI Systems
GAAT connects telemetry with policy enforcement to block violations in real time.
SQUIRE: Authoring Interactive UIs via Slot Query Intermediate Representation
**Squire** gives more precise control over AI UI generation through pointing-and-clicking and scoped affordances.
Personalized Group Relative Policy Optimization for Heterogeneous Preference Alignment
P-GRPO reduces GRPO's bias by decoupling individual preferences using per-group reward history.
ProText: A Benchmark Dataset for Measuring (Mis)Gendering in Long-Form Text
ProText is a dataset released to measure gendering and misgendering in long-form text.
Entropy-Preserving Reinforcement Learning
Controlling the entropy-reducing tendency of policy gradient training preserves exploration diversity.
Beyond Real Data: Synthetic Data from a Regularization Perspective
More synthetic data isn't always better—there is an optimal ratio between synthetic and real data.
Fewer Gaussians, More Texture: 4K Feed-Forward Textured Splatting
LGTM solves the exploding Gaussian count problem by combining textures, enabling 4K synthesis.
Athena: An Intermediate Representation for Iterative Scaffolded App Generation with LLMs
It structures LLM app generation using storyboards, data models, and GUI skeletons.
Infinitely Further: Tool-Use Unlocks Length Generalization for State Space Models
SSMs equipped with tool-use generalize beyond their length limits to problems of arbitrary length.
Drop-in Perceptual Optimization for 3D Gaussian Splatting
WD-R replaces the default loss in 3DGS, boosting both texture quality and human preference.
Revisiting Scaling Properties of Downstream Metrics in LLM Training
When the token-to-parameter ratio is fixed, downstream accuracy can also be predicted by a power law.