AI & Machine Learning

Concept erasure in text-to-image models is largely a facade that can be bypassed using text-free inversion attacks.

LLMs compute and cache confidence scores automatically during answer generation, well before they are prompted to verbalize them.

ProbeFlow achieves 14.8x faster action decoding in Vision-Language-Action (VLA) models without any retraining.

DebugLM allows developers to trace an LLM's specific behaviors back to individual training data sources.

Measuring the distance between human languages can now be done quantitatively using the attention mechanisms of multilingual transformers.

Large Language Models can maintain performance with only 16-64 unique weight values per matrix, as only the relative rank of weights matters.

Parallel multi-token prediction can be achieved in standard LLMs without training auxiliary models or modifying weights.

CARE provides a recipe for converting standard GQA models into high-efficiency Multi-head Latent Attention (MLA) architectures.

VideoAtlas enables navigation and reasoning over long-form video using compute that scales only logarithmically with video length.

Enforce formal safety and Signal Temporal Logic (STL) constraints on robotics foundation models without retraining.

MUD provides a faster, lower-overhead alternative to Muon for transformer training, achieving up to 2.6x higher throughput.

LoST introduces a semantic-first 3D tokenizer that reduces the token count for 3D shape generation by up to 99.9%.

AgentFactory shifts agent evolution from unreliable textual 'reflections' to a library of verifiable, executable Python subagents.

SkeletonLLM allows frozen Multimodal LLMs to reason about human motion by rendering skeleton sequences into their native visual modality.

DAPS++ reinterprets diffusion inverse problems as a decoupled EM-style initialization, significantly increasing restoration speed and stability.

Motion-MLLM integrates IMU egomotion data into Video-LLMs to solve the fundamental scale and spatial reasoning ambiguities of purely visual models.

Provides the first theoretical proof that Graph Transformers structurally prevent the 'oversmoothing' failure mode inherent to deep GCNs.

Imagine an AI virus that doesn't just sit there—it copies itself and jumps from one AI to the next all on its own.

A new VR headset uses mirrors to kill the lag that makes you want to puke.

These tiny sliding antennas are hacking the laws of physics to give you a perfect signal where your phone usually dies.

New AI can peer into a computer chip's microscopic guts to find "spy tech" hidden by sketchy manufacturers.

Researchers built a "ghost mode" for robots that calculates the exact path to sneak around without being seen.

Turns out the long lines at airport security were secretly keeping the whole U.S. flight network from crashing for the last decade.

RSM achieves 20x faster training for recursive reasoning models and enables test-time scaling for up to 20,000 refinement steps.

A factorial study on EHR foundation models reveals that joint encoding of code-attribute pairs (local binding) is the primary driver of performance and efficiency.

Alternating Reinforcement Learning with Rubric Rewards (ARL-RR) replaces brittle scalar reward aggregation with a semantic meta-class optimization framework.

Self-reflective program search matches or outperforms recursive language models for long-context tasks, suggesting recursion itself is not the primary driver of performance.

Dynamic Representational Circuit Breaking (DRCB) introduces an architectural defense against steganographic collusion in multi-agent RL by monitoring and shuffling latent communication bottlenecks.

Theoretical and empirical evidence suggests that the 'Key' mechanism in Attention may be redundant, proposing a 'QV' paradigm that simplifies Transformer architectures.

Atlas introduces 'Compiled Memory,' which rewrites an agent's system prompt with distilled task experience rather than using RAG or fine-tuning.

Latent Posterior Factors (LPF) bridge neural representations with structured probabilistic reasoning by converting VAE posteriors into factors for Sum-Product Networks.

Spectral Edge Dynamics (SED) provides an early-warning signal for grokking, predicting generalization up to 1,700 steps before it occurs.

Transition Flow Matching learns a global transition flow rather than local velocity fields, enabling single-step generation and transfer to arbitrary future time points.

Robot policy performance can be improved by up to 60% by identifying a single 'golden ticket' constant noise vector instead of sampling from a Gaussian.

Simulation Distillation (SimDist) enables rapid sim-to-real adaptation by transferring reward and value models directly into a latent world model.

Demonstrates that massive scaling of diverse simulator resets can replace manual curriculum engineering for complex dexterous manipulation.

Reduces high-quality 3D head avatar creation time from over 24 hours to 0.5 seconds per frame.

Reveals that models with identical predictive performance produce fundamentally different feature attributions based solely on their hypothesis class.

Introduces a privacy-preserving ML framework that achieves strong non-invertibility without the utility loss of Differential Privacy or the cost of Homomorphic Encryption.

Fuses categorical sampling into the LM-head matmul to eliminate logit materialization and speed up LLM decoding by up to 19%.

Analyses over 10,000 experiments to prove that LLM agents are capable of genuine architectural discovery rather than just hyperparameter tuning.

Provides empirical evidence that structural sparsity in Vision Transformers does not lead to improved semantic interpretability.

Demonstrates a complete AI-assisted mathematical research loop where a mathematician wrote zero lines of formal code to verify complex physics equilibria.

Integrates LLM agents with the industry-standard Rosetta software to automate physics-based protein design for non-canonical amino acids.

Releases 70B parameter models that operate entirely on bytes, effectively 'liberating' LLMs from static tokenizers.

Derives closed-form power-law scaling for hyperparameters like learning rate and batch size using modern optimization theory rather than expensive empirical sweeps.

Introduces per-token adapter routing, allowing a single sequence to dynamically utilize multiple specialized LoRA experts.

Provides the first formal proof that safety is non-compositional, meaning two individually safe AI agents can become hazardous when combined.

Enables the prediction of an adapter's task, performance, and attributes directly from its LoRA weights without any inference or data access.

Finds that filtering knowledge at 'write-time' (ingestion) maintains 100% RAG accuracy under noise levels where standard 'read-time' filtering completely collapses.