MIT CSAIL

Self-distillation methods fail to generalize reasoning gains to unseen problems and their gradients offer weak policy improvement signals, challenging their efficacy beyond in-distribution imitation.

AI agents can now build and reuse persistent, artifact-specific knowledge, dramatically improving their ability to find relevant information for complex tasks and reducing operational costs.

A simple visual harness transforms a general multimodal model into a perfect reasoner on ARC-AGI-3, completing all games with drastically fewer actions than humans.

Most biological reasoning models fail to utilize their biological foundation model inputs, instead relying on textual data even when presented with conflicting biological information.

It is shown that the workspace token can be used as a drop-in replacement for observations during deployment, enabling policies to solve memory-intensive tasks without the need for VLM reasoning in-the-loop.

SVI-DAG outperforms existing Bayesian methods by effectively quantifying uncertainty in causal inference while leveraging prior knowledge and edge dependencies.

Data synergy can either amplify or diminish model performance, revealing that the right dataset combinations are crucial for optimal language model training.

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.

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

LLMs reveal a modular cognitive architecture strikingly similar to that of the human brain, challenging our understanding of intelligence across different systems.

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

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

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

Adaptive opponents can significantly alter the landscape of regret minimization, and our new RP-Regret metric reveals how to achieve better cooperative outcomes in repeated games.

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.

Forget brute-force scaling: crafting the *right* context from past experiences unlocks surprisingly large gains in LLM agent performance.

Particle filter models of sentence processing inherently predict "digging-in" effects—where disambiguation difficulty increases with the length of the ambiguous region—a phenomenon not captured by surprisal-based models.

HybridRAG-Bench reveals that existing benchmarks overestimate the reasoning abilities of retrieval-augmented LLMs due to contamination, offering a more realistic evaluation using up-to-date scientific knowledge.

Achieving 80% accuracy on VQA v2.0 proves that combining Visual BERT, ViLT, and memory-augmented attention can significantly outperform traditional VQA models.