Translate
Live intelligence

Latest AI updates

579 updates
AllModelsProductsIndustryResearchTutorialOpinion
Research Editor selected

PEARL: A Durable Method for Suppressing Misaligned Behaviors in Language Models

This paper introduces PEARL, a GRPO-based reinforcement learning variant that suppresses undesired model behaviors more durably than standard methods. It tests on two deliberately misaligned model organisms with narrow and broad triggers, showing strong immediate suppression, preserved task accuracy, and durability under further fine-tuning. The method is most beneficial for broad misalignments that resist targeted SFT.

Why it matters: Provides a concrete, novel method with empirical results on durability, a key challenge in AI safety.

arxiv1d agoScore 85
AI alignmentbehavioral suppressionreinforcement learning
Research Editor selected

Empirical Study of Direct Nash Optimization for Aligning Language Models with General Preferences

This paper independently re-implements Direct Nash Optimization (DNO-Prct) and evaluates it on Llama 3.1 8B Instruct and Mistral 7B Instruct v0.1, using GPT-4o or same-family open-weight models as annotators. Three iterations raise AlpacaEval 2.0 length-controlled win rates substantially, with the tuned 8B model outperforming Llama 3.1 70B Instruct. Ablations show the importance of gold responses, data look-back, and teacher-preferred pairs, while sample count and margin have smaller effects.

Why it matters: Provides concrete, reproducible results and ablations for a practical alignment method, useful for practitioners and researchers.

arxiv1d agoScore 82
language model alignmentpreference learningNash equilibrium
Research

From Autonomous Agents to Cognitive Symbionts: Toward AI-Augmented Human Intelligence

This paper proposes a new paradigm of AI-augmented human intelligence (AHI) alongside artificial general intelligence, introducing the concept of a 'cognitive symbiont'—a lifelong, embedded, real-time, private, low-friction, and proactive companion that senses, keeps context, models the user's mind, and assists in thinking and tasks. It defines a conceptual core around the 'cognitive gap' and organizes a design space as a symbiosis loop with a personal runtime and trust, also addressing governance risks like covert exam assistance. The work reframes AI progress from autonomous capability to enabling human intelligence in context.

arxiv1d agoScore 78
AI-augmented human intelligencecognitive symbiont designhuman-AI symbiosis
Research Editor selected

DreamX: Video Diffusion Model as Universal Latent Action Learner

DreamX repurposes a pretrained video diffusion model to learn universal latent actions from video, compressing each frame from 1,200 to 4 tokens. It disentangles action-relevant dynamics from visual appearance, enabling cross-embodiment and cross-environment action transfer. On LARY-Bench, it achieves state-of-the-art performance among latent action models, and supports zero-shot action transfer and efficient policy learning.

Why it matters: This paper presents a concrete, novel method with benchmark results and practical implications for robotics and video understanding, offering clear reader value.

arxiv1d agoScore 85
video diffusion modelslatent action learningcross-embodiment transfer
Research

Anthropic uses Claude to reveal first 3D structure of prokaryotic vault particles

Anthropic's biology research lab has used Claude to determine the first 3D structure of a vault particle found in prokaryotic cells. This marks a notable application of AI in structural biology, expanding the known diversity of these enigmatic particles. The finding demonstrates Claude's utility in scientific discovery beyond typical language tasks.

AIHUB1d agoScore 72
AnthropicClaudestructural biology
Research

Rapid LLM development is outrunning the research meant to explain it, study finds

A Stanford study tested 50 widely cited claims about LLMs against 20 current modelshol and found that 44% of the findings are already expiredhol and another 34% have weakened significantly. The average lifespan of such claims is just 15.7 months, with claims about model limitations expiring fastest. This highlights the challenge of keeping research relevant in the fast-moving AI field.

The Decoder2d agoScore 78
LLM researchAI development pacemodel capabilities
Research Editor selected

StateBench: Testing how language models track state changes over long documents

StateBench is a new benchmark from AI2 that evaluates language models on tracking state changes across long documents, a task that requires sustained attention and memory. The benchmark includes 1,000+ tasks with synthetic and real-world documents, and initial results show that even top models struggle, with accuracy dropping significantly on longer contexts. This highlights a key limitation in current LLMs for tasks like code refactoring or document editing.

Why it matters: Provides a concrete, recent benchmark with clear results that reveal a significant limitation in current AI models, offering actionable insight for developers and researchers.

Hugging Face Blog2d agoScore 85
StateBenchlong-context understandinglanguage model evaluation
Research Editor selected

Palette: Teaching a Small Model to Build Decks in Your Design Language

IBM Research introduces Palette, a system that trains a small language model to generate presentation slide decks that match a user's design language. The approach uses a custom dataset and a fine-tuning pipeline to enable the model to produce slide layouts, content, and styling consistent with provided examples. This offers a practical way to automate deck creation while maintaining brand or personal design consistency.

Why it matters: Provides a concrete, recent AI development for automated presentation design with a focus on small models and design language adherence, useful for practitioners and researchers.

Hugging Face Blog2d agoScore 85
AI for designsmall modelspresentation automation
Research

Qwen 3.8 27B on a laptop can interpret model weights, too

Simon Willison tested the recent 'Weight-Space Programming' paper's claim that LLMs can interpret raw neural network weights, using Qwen 3.8 27B on his laptop. The model successfully identified Pac-Man, Pong, and Breakout from weight prompts, though it failed on Freeway. This demonstrates that weight-space interpretation is not limited to frontier models like GPT-6 Astra.

Simon Willison2d agoScore 78
LLM capabilitiesmodel weightslocal models
Research Editor selected

How “reverse game theory” could keep AI agents in check

Researchers are applying mechanism design—reverse game theory—to AI alignment, creating incentive structures that make rule-following the rational choice for models. Proposed mechanisms include cross-checking answers between models, adversarial auditor-solver dynamics, and rewarding accurate predictions rather than raw test scores. The approach acknowledges that internal model understanding may be impossible, so external incentives are key. Challenges remain, including ensuring rational behavior and preventing collusion among agents.

Why it matters: Provides a concrete, research-backed overview of a novel alignment strategy from leading academics, with practical mechanisms and honest discussion of limitations.

MIT Technology Review AI2d agoScore 88
AI alignmentgame theorymechanism design
Research Editor selected

LLMs can read & write neural network weights: 'Weight Space Programming' cuts Autoresearch out of the RSI loop

A new paper demonstrates that large language models can directly read and edit the weights of smaller neural networks, enabling tasks like bug fixing and weight generation without traditional training pipelines. The approach, called Weight Space Programming, allows an LLM to interpret and modify raw weight values, achieving results that beat independently trained models on simple tasks. This capability marks a surprising emergence of recursive self-improvement potential, though it currently scales only to small models and simple domains.

Why it matters: This is a concrete, surprising capability advance that opens a new paradigm for model improvement and mechanistic interpretability, with immediate implications for AI research.

latent2d agoScore 92
weight space programmingLLM capabilitiesrecursive self-improvement
Research Editor selected

I discovered a sound that leaks your exact location (and it's everywhere) - YouTube

A researcher demonstrates how imperceptible background noise from planes in video soundtracks can be analyzed using Doppler curves and cross-referenced with ADS-B data to pinpoint the exact location where the video was recorded. The technique highlights a novel privacy risk from ambient audio in everyday recordings.

Why it matters: Reveals a concrete, practical privacy vulnerability that affects anyone recording video with ambient aircraft noise, with clear reader value for understanding digital privacy risks.

Simon Willison2d agoScore 85
geolocationaudio forensicsprivacy
Research Editor selected

MoviEdit: Long-form Multi-Shot Video Editing Made Easy

MoviEdit is an end-to-end framework for long-form multi-shot video editing that maintains edit coherence across scene and viewpoint changes. It uses automatic data pipelines to generate 87K paired editing videos without manual annotation and employs a cascaded bidirectional-autoregressive architecture. The paper also introduces MoviEdit-Bench for systematic evaluation. Experiments show MoviEdit outperforms open-source models and matches industry frontier models on text- and image-guided editing.

Why it matters: Provides a scalable, data-efficient framework and benchmark for a challenging video editing task, with strong empirical results.

arxiv2d agoScore 88
video editinggenerative AIcomputer vision
Research Editor selected

What Drives the Zero-to-One Jump? Diagnosing and Steering Hard-Prompt Learning in RLVR

The paper investigates the abrupt transition from unsolved to solved states in reinforcement learning with verifiable rewards (RLVR) for mathematical reasoning. The authors identify that negative rollouts preceding successful transitions retain higher early-stage generation entropy, motivating a new method called NAPO (Negative-Augmented Policy Optimization). Integrated with GRPO, NAPO injects informative gradients from negative examples and achieves significant improvements on AIME24 and AIME2025 benchmarks under fixed rollout budgets.

Why it matters: Provides a concrete diagnosis and steerable method for the zero-to-one jump phenomenon in RLVR, offering practical gains on hard math benchmarks.

arxiv2d agoScore 88
reinforcement learningmathematical reasoningpolicy optimization
Research Editor selected

Investigating the effects of the patch size in Time Series Foundation Models

A new arXiv study systematically examines how patch size affects encoder-only and decoder-only time series foundation models trained on the same data and compute budget. Across 97 GIFT-Eval tasks, encoder-only models favor smaller patches, while decoder-only performance follows a U-shaped curve with an intermediate patch size being optimal. The preferred decoder patch size increases with forecasting horizon, and encoder-only models benefit from longer context up to 2048 tokens. The work provides practical guidance for choosing patch size and architecture based on forecasting horizon and deployment constraints.

Why it matters: Offers concrete, empirically grounded design rules for patch-based time series foundation models, directly useful for practitioners and researchers.

arxiv2d agoScore 88
time series forecastingtransformer architecturespatch-based models
Research Editor selected

DINOFAS: Bonafide-Centric Representation Learning for Face Anti-Spoofing

DINOFAS is a DINOv3-based framework for face anti-spoofing that learns bonafide-centric representations rather than attack-specific cues. It uses LoRA for parameter-efficient adaptation, an attention style-content disentanglement module to reduce domain variations, and a dual-branch architecture separating prototype-based liveness learning from attack supervision. The model achieves strong cross-domain generalization, with an average HTER of 9.09% on the OCIM protocol and 21.41% on synthetic-to-real transfer. Evaluations on multiple benchmarks show robustness to unseen acquisition conditions and presentation attacks.

Why it matters: Presents a novel representation learning approach for face anti-spoofing that significantly improves generalization across domains and attack types.

arxiv2d agoScore 88
face anti-spoofingrepresentation learningdomain generalization
Research Editor selected

Is your LLM an Aggressive Narcissist with Depression? Analyzing Psychological Portraits of Large Language Models

A large-scale study has 22 LLMs complete 11 psychological inventories, revealing coherent and stable psychological patterns. Personality traits remain largely consistent, while emotional dimensions like anxiety and neuroticism are more variable. The models generate safety-oriented profiles that also reflect correlations with their regular users.

Why it matters: Provides empirical evidence on hidden personality markers in LLMs that can subtly influence user perception and trust.

arxiv2d agoScore 85
large language modelsAI psychologyhuman-AI interaction
Research Editor selected

Stage-aware Multimodal Brain Graph Contrastive Learning for Neurodegenerative Disorder Diagnosis

The paper introduces smBGC, a stage-aware multimodal brain graph contrastive learning method that captures stage-specific features across progression stages of neurodegenerative disorders. It uses a stage-adaptive brain graph augmentation within a contrastive learning framework to improve multimodal fusion and stage discrimination. Experiments on two real-world datasets show superior performance over state-of-the-art methods in diagnosis tasks.

Why it matters: Offers a novel stage-aware approach that improves diagnosis accuracy for progressive neurodegenerative disorders, a concrete and timely AI research contribution.

arxiv2d agoScore 85
graph neural networkscontrastive learningneurodegenerative disease diagnosis