Multi-agent LLM systems are increasingly deployed as autonomous collaborators, where agents interact freely rather than execute fixed, pre-specified workflows. In such settings, effective coordination cannot be fully designed in advance and must instead emerge through interaction. However, most prior work enforces coordination through fixed roles, workflows, or aggregation rules, leaving open the question of how well self-organizing teams perform when coordination is unconstrained. Drawing on organizational psychology, we study whether self-organizing LLM teams achieve strong synergy, where…
The problem of domain generalization concerns learning predictive models that are robust to distribution shifts when deployed in new, previously unseen environments. Existing methods typically require labeled data from multiple training environments, limiting their applicability when labeled data are scarce. In this work, we study domain generalization in an anti-causal setting, where the outcome causes the observed covariates. Under this structure, environment perturbations that affect the covariates do not propagate to the outcome, which motivates regularizing the model’s sensitivity to…
Reasoning Large Language Models (LLMs) enable test-time scaling, with dataset-level accuracy improving as the token budget increases, motivating adaptive reasoning—spending tokens when they improve reliability and stopping early when additional computation is unlikely to help. However, setting the token budget, as well as the threshold for adaptive reasoning, is a practical challenge that entails a fundamental risk-accuracy trade-off. We re-frame the budget setting problem as risk control, limiting the error rate while minimizing compute. Our framework introduces an upper threshold that stops…
Diffusion Large Language Models (dLLMs) have emerged as a promising alternative to purely autoregressive language models because they can decode multiple tokens in parallel. However, state-of-the-art block-wise dLLMs rely on a “remasking” mechanism that decodes only the most confident tokens and discards the rest, effectively wasting computation. We demonstrate that recycling computation from the discarded tokens is beneficial, as these tokens retain contextual information useful for subsequent decoding iterations. In light of this, we propose Residual Context Diffusion (RCD), a module that…
Diffusion (Large) Language Models (dLLMs) now match the downstream performance of their autoregressive counterparts on many tasks, while holding the promise of being more efficient during inference. One critical design aspect of dLLMs is the sampling procedure that selects which tokens to unmask at each diffusion step. Indeed, recent work has found that heuristic strategies such as confidence thresholding improve both sample quality and token throughput compared to random unmasking. However, such heuristics have downsides: they require manual tuning, and we observe that their performance…
Large language models can exhibit emergent reasoning behaviors, often manifested as recurring lexical patterns (e.g., “wait,” indicating verification). However, complex reasoning trajectories remain sparse in unconstrained sampling, and standard RL often fails to guarantee the acquisition of diverse reasoning behaviors. We propose a systematic discovery and reinforcement of diverse reasoning patterns through structured reasoning, a paradigm that requires targeted exploration of specific reasoning patterns during the RL process. To this end, we propose Ctrl-R, a framework for learning…
Understanding how transformer components operate in LLMs is important, as it is at the core of recent technological advances in artificial intelligence. In this work, we revisit the challenges associated with interpretability of feed-forward modules (FFNs) and propose MemoryLLM, which aims to decouple FFNs from self-attention and enables us to study the decoupled FFNs as context-free token-wise neural retrieval memory. In detail, we investigate how input tokens access memory locations within FFN parameters and the importance of FFN memory across different downstream tasks. MemoryLLM achieves…
Reinforcement learning (RL) finetuning has become a key technique for enhancing large language models (LLMs) on reasoning-intensive tasks, motivating its extension to vision language models (VLMs). While RL-tuned VLMs improve on visual reasoning benchmarks, they remain vulnerable to weak visual grounding, hallucinations, and over-reliance on textual cues. We show that simple, controlled textual perturbations—misleading captions or incorrect chain-of-thought (CoT) traces—cause substantial drops in robustness and confidence, and that these effects are more pronounced when CoT consistency is…
Maximum inner product search (MIPS) is a crucial subroutine in machine learning, requiring the identification of a vector taken within a database (the keys) that best aligns with a given query. We propose amortized MIPS: a regression-based approach that trains neural networks to directly predict MIPS solutions, amortizing the cost of repeatedly solving MIPS for queries drawn from a known distribution over a fixed key database. Our key insight is that the MIPS value function is the support function of the set of keys, a well-studied convex function whose gradient yields the optimal key. This…
Visual tokenizers map high-dimensional raw pixels into a compressed representation for downstream modeling. Beyond compression, tokenizers dictate what information is preserved and how it is organized. A de facto standard approach to video tokenization is to represent a video as a spatiotemporal 3D grid of tokens, each capturing the corresponding local information in the original signal. This requires the downstream model that consumes the tokens, e.g., a text-to-video model, to learn to predict all low-level details “pixel-by-pixel” irrespective of the video’s inherent complexity, leading to…
LLM-as-a-judge panels aggregate votes from multiple models, with the expectation that diverse models yield more reliable evaluations. We develop a framework to measure the true informational value of such panels and quantify how far their reliability falls short of the independent-voting ideal. Testing a panel of 9 frontier LLMs from 7 model families on three natural language inference datasets (each with 100 human annotations per item), we find that the 9 judges effectively provide only about 2 independent votes’ worth of information. Roughly three-quarters of the panel’s nominal independence…
When annotators disagree on a label, the disagreement itself carries signal—and the number of annotators needed to capture it depends on the evaluation metric. We fine-tune NLI models on label distributions subsampled from ChaosNLI, a dataset providing 100 independent annotator judgments per item, and identify metric-dependent saturation. In our 3-class NLI setting, entropy correlation—whether the model identifies which items elicit disagreement—requires N ≈ 20–50 annotators to converge, while distributional match (KL divergence) saturates by N ≈ 10 (87–95% of improvement across five model…
Splitting the “separation kernel” off from the rest of the Nitro security system and using only a subset of the Rust programming language to code it enabled its formal verification.
A new chiplet architecture, custom die-to-die connectivity, and support for DDR5-8800 memory and the latest PCIe gen6 interconnects improve performance by 25% for general-purpose and agentic AI workloads.
The harnesses that mediate between models and tools in agentic systems are becoming their own performance bottleneck, but a few simple design principles can fix what ails them.
Our next generation of Apple Intelligence is centered around our users, integrated deeply into our operating systems, and powered by a bold new architecture with privacy at its core. At the heart of this architecture is our third generation of Apple Foundation Models (AFM), a family of five foundation models custom-built in collaboration with Google. These span from on-device models to server-based models running on Private Cloud Compute. Apple Foundation Models are built to unlock a wide range of helpful experiences for our users, like an entirely new Siri and intelligent tools that make…
—Introduction of various research papers on new models, including a neuro-symbolic framework for counterfactual explanations and discrete diffusion language models for interactive text generation.
—Development of Object Aligner for optimizing LLM prompts and enhancing JSON schema similarity scores.
Research
—Multiple papers addressing challenges in reinforcement learning, programming by example, and multi-agent systems for improved forecasting.
—Studies on multilingual TTS evaluation, Bayesian learning for hardware impairments, and the robustness of programming by example systems.
Tools
—New GitHub repositories like promptdiff for version control of LLM prompts and agent-replay for debugging AI agent execution.
—Agents-control-tower for monitoring multiple AI agents in a single terminal environment.
Discussion
—Japan's top court ruling that AI cannot be listed as an inventor on patent applications sparked significant online discussion.
—Ongoing debates about the implications of LLMs watching videos and the absence of LLM code in dependencies, highlighting concerns about AI safety and transparency.