The study shows that best-of-N alignment closely approaches the theoretical maximum reward gain, while PPO and GRPO remain substantially suboptimal.
The method learns query-dependent weights to compose cached state encodings, improving performance on document QA, needle-in-a-haystack retrieval, and tool use.
The study demonstrates that gating mechanisms in SSMs cause models to learn in-weights memorization solutions, delaying or preventing convergence to correct in-context learning even without architectural memory limits.
Study shows judge quality varies by task and criteria, with paired comparisons outperforming independent scoring and top models exhibiting exaggerated sensitivity to subtle details.
The technique trains small neural networks offline to route and rewrite queries, achieving up to 8x faster search and higher recall than centroid-based indexing on the Natural Questions dataset.
The framework decouples planning and grounding agents, using dense rewards and PPO to outperform Gemini-3.5-Flash on real iOS applications.
The method achieves state-of-the-art quality on the Ava-256 benchmark while using 8x fewer Gaussians.
The study shows a complexity-dependent crossover where Plan-then-Retrieve excels on multi-tool queries while Retrieve-then-Plan wins on single-tool tasks.
SSR reduces per-turn reasoning latency by over 90% and total inference latency by 28–54% on 2B and 4B models.
A cluster of four M5 Ultra Mac Studios achieves over 36 tokens per second on the 816B-parameter Muse Spark Open model using new RDMA-based distributed inference.
The method achieves a mean F1 score of 0.798 across 14 language pairs in CS-FLEURS using the Whisper-small encoder.
The study presents the first tri-modal Masked Diffusion Models pretrained on text, image-text, and audio-text data, achieving strong results at 3B scale.
Mixing high-resource auxiliary data boosts performance equivalent to 2–3× more unique target data on validation loss.
The method improves math performance by up to 18.1 points and agentic search by 5–9 points while reducing token costs by 30–70%.
The system aggregates idle GPUs from macOS devices to execute large-scale machine learning inference without new hardware procurement.
The study evaluated LLMs against human annotators across 70 language pairs using MQM and ESA schemes, finding both unreliable in most settings.
Apple researchers introduced Normalizing Trajectory Models (NTM), a framework that matches strong baselines in four sampling steps while retaining exact likelihood over the generative trajectory.
RISED uses LLM-generated rubrics to guide data selection and self-distillation, achieving the highest mean pass rate across diverse interactive environments.
The study identifies four key sites where users negotiate ontological boundaries when defining their own phenomena for machine learning training.
Apple research shows confidence-based sampling in discrete diffusion models selects dependent token groups, causing a 29x deviation from the training distribution despite perfect per-sample metrics.
Interventions like auxiliary classifiers reduced phone discrimination error from 11.6% to 10.4% in English/French HuBERT models.
Apple researchers introduce RLTL;DR, a method that boosts Pass@1 scores from near-zero to 12–13% on difficult tasks by internalizing self-generated feedback.
A study shows that under equal time budgets and the same LLM backbone, elaborate multi-agent harnesses offer no performance advantage over minimal-harness coding agents for autonomous ML engineering.
SCLATE compresses month-long agent scenarios into hours using a hybrid simulated clock and open event scheduler.
AI Briefing