MIT CSAIL

Deployment rules can shift multi-agent AI outcomes dramatically, with fatality rates varying by up to 58 percentage points based solely on the chosen rule.

Frequency usage in transformers is not random; it’s intricately tied to the data’s dependency structure, revealing a data-driven mechanism behind RoPE's emergent behavior.

Silent policy violations in tool-using LLMs can be mitigated by deterministic gates, improving success rates by over 12 percentage points in critical tasks.

Multimodal unlearning could revolutionize how we handle sensitive data in AI, enabling targeted removal without sacrificing model performance.

LLMs can be trained to negotiate like expert agents, extracting significantly higher surpluses by strategically exploring buyer markets rather than fixating on immediate bids.

Nearly 80% of AI-generated pull requests are submitted concurrently, raising critical questions about collaboration efficiency and merge conflicts in AI coding agents.

The dominance of a few countries and institutions in AI bias research risks creating a narrow lens through which fairness is defined and addressed.

LLM-driven program synthesis can automate EEG feature engineering while ensuring interpretability and high detection accuracy.

Low diversity in training data can lead to substantial performance drops in language models, revealing a critical oversight in data augmentation practices.

Projected reads in PatchOptic not only cut token costs but also ensure that local updates remain valid in the context of shared-state workflows.

DiscoPER not only automates hypothesis generation but also self-analyzes its discoveries, revealing hidden patterns and expanding the search space in unprecedented ways.

The choice of performance metrics could determine whether AI capabilities remain concentrated among the wealthy or proliferate across a broader developer base.

NNPs can achieve near-chemical accuracy in enzyme catalysis predictions with less than 1,000 system-specific data points, revolutionizing the efficiency of mechanistic studies.

Adversarial training with human demonstrations can significantly enhance the quality and diversity of language model outputs while preserving accuracy.

Fixed counterfactual explanations can lead LMs to generate more accurate introspections about their behaviors, even as those behaviors change over time.

SLIM-RL achieves state-of-the-art performance on math and code tasks with nearly half the training samples required by traditional trajectory-aware methods.

Automating freeway network extraction from OSM can cut analyst effort by two-thirds, making large-scale freeway simulations feasible.

Historical failure records can be transformed into diverse testing scenarios for autonomous driving, revealing critical system vulnerabilities with minimal effort.

Relying on text-based rationales for dementia classification can actually degrade performance, revealing the need for more sophisticated approaches like DeTAiL.

Evaluating creative AI requires recognizing that professional disagreement reflects genuine taste differences, not just noise in measurement.

A unified platform for molecular machine learning that supports 100 elements and incorporates uncertainty quantification could democratize access to advanced chemical property predictions.

Generative AI agents can reveal how personalization algorithms amplify toxic content in ways that vary dramatically by user ideology.

ArBG achieves a remarkable 60% reduction in zero-shot energy error for peptide systems, challenging the dominance of flow-based sampling methods.

FracEvent achieves superior event timing and downstream performance by accurately modeling pixel dynamics, outperforming traditional simulators.

Memory consistency in video generation models falters significantly when objects disappear, with state-of-the-art models struggling to recover updated states upon reappearance.

Dynamic frame rates in audio autoencoders can drastically enhance efficiency, allowing for smarter resource allocation in neural compression tasks.

GPUSparse achieves a staggering 235x speedup over traditional CPU methods while maintaining exact scoring, revolutionizing real-time retrieval efficiency.

Achieving a staggering 220x speedup in MaxSim scoring while preserving exact retrieval quality could revolutionize the efficiency of multi-vector retrieval systems.

Complex manipulation capabilities can be achieved by dynamically composing simple behaviors, leading to unprecedented precision and adaptability in real-world tasks.

Cross-lingual exploration can unlock hidden knowledge in LLMs, improving factual recall and consistency across 17 languages.

Current evaluation metrics for AI-powered AAC systems overlook the intersectional nuances of user needs, risking ineffective communication solutions.

Fixed exponents in neural scaling laws reveal that optimizing coefficients could unlock significant performance gains in large language models.

Turn-final words are not just longer; they provide a crucial prosodic cue for predicting conversational turns, localized mainly in the final syllable.

Over 200,000 synthesized words reveal the intricate relationship between articulatory gestures and acoustic landmarks in speech.

Gaussian soft labels boost landmark detection performance by 7% over traditional hard labels, revealing the importance of modeling annotation variability.

Decentralized traffic management for autonomous aircraft can achieve high performance without centralized coordination, adapting seamlessly to complex environments.

Balancing productivity and stability reveals that stronger synchronization can paradoxically increase systemic fragility in multi-agent systems.

Trajectory mining reveals skill structures but fails to translate these insights into meaningful performance gains for downstream policies.

Turing-RL reveals that training user simulators for indistinguishability can dramatically improve their performance in simulating human interactions.

WalkOCC achieves superior sidewalk occupancy prediction by leveraging unpaired monocular images, eliminating the need for costly 3D annotations.

Machine learning can transform 2DES by extracting maximum insights from limited data while guiding experimental design for improved accuracy.

Training climate emulators on a single optimized scenario can outperform those trained on six standard pathways, challenging the notion that more data always leads to better performance.

Executable programs can now replace attention heads in transformers with minimal performance loss, achieving over 75% similarity to original patterns.

The largest-ever verification campaign for Rust's standard library reveals significant vulnerabilities in unsafe code, underscoring the need for robust static verification methods.

Nonuniform width allocation in transformers can lead to a 22% reduction in FLOPs while enhancing language modeling performance.

MixTIME reveals that integrating diverse image modalities can significantly enhance the precision of immune biomarker predictions in oncology, outperforming traditional single-modality approaches.

Post-congestion pricing, NYC saw a surge in transit ridership while overall travel demand fell, highlighting the complex interplay of urban mobility and pricing policies.

Retaining visual figures in skill artifacts boosts CUA performance by over 23 points, proving that seeing is believing in agent training.

Coordination-free DIDs could revolutionize decentralized identity management by enabling instantaneous updates without the overhead of global ordering.

Dynestyx transforms the landscape of Bayesian analysis by making advanced state-space modeling techniques accessible to practitioners through a unified interface.

LLMs show significant variability in the actionability of their UX critiques, with some models outperforming others across different product categories.

Naively scaling context length in imitation learning is surprisingly robust, challenging previous assumptions about brittleness in policy performance.

LLMs can now interpret quantum operations, achieving state-of-the-art results in circuit synthesis while allowing natural language constraints.

Current AI agents excel in structured tasks but falter at generating novel insights and tackling open-ended scientific challenges.

Optimizing input configurations can boost LLM performance in pathology tasks, closing the gap with specialized models and challenging assumptions about domain-specific training.

Even state-of-the-art multimodal models struggle with reliability in clinical tool use, revealing critical gaps in AI agent performance.

The "curse of precision" reveals how reliance on AI-generated content can degrade model performance by homogenizing training data.

Action-chunking policies can lead to premature robot assistance, but a novel steering method effectively mitigates this issue, enhancing collaboration efficiency.

Standard stereo methods can produce 3D models of Martian terrain, but achieving reliable reconstruction demands careful consideration of domain-specific challenges.

State inertia in full-duplex spoken language models can lead to missed user input, but activation steering effectively mitigates this issue, boosting comprehension rates significantly.

Gradient inference can revolutionize how we tackle gradient estimation in complex probabilistic programs, enabling new state-of-the-art estimators that outperform traditional methods.

Despite promising engagement benefits, foundation model-based care robots struggle with reliability and lack robust evidence for clinical impact.

Systematic gaps in AI evaluation reporting are exposed, revealing inconsistencies that hinder reliable comparisons across thousands of models and benchmarks.

Current clinical AI systems often neglect the temporal dimension of patient care, limiting their effectiveness in longitudinal reasoning.

AI systems currently miss critical temporal and interpretive elements of clinical reasoning, limiting their effectiveness in real-world healthcare settings.

Design choices in agent memory systems can significantly shift operational costs, revealing critical trade-offs that impact long-horizon task performance.

Meridian achieves accurate global localization in unstructured environments without the need for area-specific training, outperforming traditional methods.

Monte Carlo methods can now compute Steklov spectra orders of magnitude faster while handling complex, disconnected geometries in real-world datasets.

USAD 2.0 achieves state-of-the-art audio understanding by seamlessly integrating self-supervised and supervised learning techniques, scaling to one billion parameters.

Active exploration can dramatically enhance adults' ability to reason about complex causal relationships, but even with this advantage, they still struggle compared to simpler tasks.

Success in long-horizon tasks hinges more on an agent's iterative persistence than on the quality of its initial solution.

Text-to-image models may only need basic word meanings and order, not complex contextual embeddings, to produce high-quality images.

SeClaw reveals that existing benchmarks fall short in capturing the complexities of agent behavior, enabling a more nuanced evaluation of security risks in autonomous systems.

Voice recordings can reveal the oscillating states of Recurrent Respiratory Papillomatosis, providing a unique longitudinal perspective on a rare laryngeal disease.

LRMs may ace problem-solving but falter dramatically in reasoning evaluation, scoring only 48% on flawed solutions that humans handle with ease.

Multimodal pretraining doesn't guarantee better alignment with human reading patterns, suggesting that language-internal representations are still king when modeling how humans process text.

Generating synthetic data for humanoid robots can boost loco-manipulation performance by 20% compared to relying solely on real-world data.

Unconditional image diffusion models can now perform continuous super-resolution without task-specific architectures or retraining, simply by varying the starting timestep.

Finally, a tactile pose estimation method that nails yaw tracking, unlocking more precise and robust robotic manipulation.

Subword tokenization just got a whole lot more efficient: ToaST slashes token counts by 11% and boosts language model performance by up to 7.6% compared to standard methods.

You can now audit Rényi differential privacy with near-optimal sample complexity, thanks to a new framework that directly estimates Rényi divergence using Donsker-Varadhan estimators.

Matching the full posterior covariance in Gaussian DDPMs slashes path KL error and unlocks faster, higher-quality sampling with a surprisingly simple Lanczos-based method.

Achieve 9.97% higher accuracy in cross-domain human activity recognition while simultaneously reducing computation by 6.4x with a new sensor data tokenization and attention mechanism.

Coordinating AI agents across scientific disciplines only boosts performance when each discipline captures a unique piece of the puzzle, otherwise, simpler combined summaries often suffice.

LLMs trained with Vector Policy Optimization (VPO) learn to produce diverse solutions that unlock previously unsolvable problems in evolutionary search, outperforming models optimized for single scalar rewards.

LMs encode grammaticality as a distinct feature in their hidden representations, separable from raw string probability and generalizable across languages.

Current alignment benchmarks are misleading: even if a model aces them, its real-world alignment could be totally different depending on the specific deployment context.

Imagine a workspace that subtly shifts lighting and sound to match your mood, all powered by an LLM that understands your needs – this paper explores the potential and pitfalls of that reality.

Quantum kernels unlock signal in medical image embeddings where classical methods fail, suggesting a new path for extracting value from medical foundation models.

Automated identification of individual animals can only be effective if it aligns with ecological questions and data practices, not just algorithmic accuracy.

Imagine slashing the human effort needed to go from hypothesis to submission-ready ML theory paper by orders of magnitude.

Forget complex fixed-point machinery: this work offers a dramatically simpler and more efficient route from external regret to $Φ$-regret minimization.

Multi-event video generation gets a 33% quality boost with TS-Attn, a training-free attention mechanism that dynamically aligns video content with complex temporal prompts.

Cyclic equalizability, a concept relevant to card-based cryptography, boils down to having identical Parikh vectors.

ZKP proving, previously bottlenecked by MSM and NTT operations, can now achieve up to 10x higher throughput on TPUs thanks to a novel framework that reformulates ZKP kernels for AI-ASIC execution.

Even the best LLMs still stumble on Olympiad-level math, and retrieval quality is the bottleneck for retrieval-augmented problem solving, according to the new MathNet benchmark.

LLMs may *look* collaborative, but the reality is often a fragile dance of misunderstandings and repairs because the interaction lacks sufficient "grounding."

Uncover the hidden assumptions baked into LLM responses with a new interactive system that lets you explore alternative conceptual framings and values.

Users feel more creative and in control when building images step-by-step from sketches, rather than wrestling with a one-shot text-to-image generator's fully-formed (and often unwanted) details.

Multi-agent systems can find 5x more real-world events in satellite imagery than traditional methods, unlocking a wealth of training data for multi-temporal change detection.