PSBC and Huawei convert existing GQA-based LLMs to MLA architecture, cutting KV cache by 72% and more than doubling concurrent users in production financial deployments
A new post-hoc framework uses causal inference to scrub toxic audio from retrieval results without retraining the underlying model or mangling the original query.
ProSPy combines SQL data retrieval with Python analysis in a four-stage agentic pipeline, hitting 60% accuracy on the notoriously hard Spider 2.0 enterprise benchmark
A new benchmark exposes that frontier multimodal AI models cannot make truly random choices, collapsing to a single option up to 97% of the time even when explicitly told to pick randomly.
TARPO teaches LLMs to dynamically switch between thinking in words and thinking in raw vector space, solving a core RL exploration problem in latent reasoning
Membrane's contrastive memory pairs every blocked attack with a permitted benign twin, slashing false refusals while topping safety benchmarks across six jailbreak types
A new geometric theory explains why LLM steering vectors randomly succeed or fail, replacing the flawed Linear Representation Hypothesis with a cylindrical model of concept space
AdaPLD upgrades model-free speculative decoding with semantic retrieval and branched draft trees, hitting 3.10x speedup without any extra models or training.
A new estimator beats Double Machine Learning's error rate for causal inference, eliminating first-order noise without requiring extra model assumptions
LatentWave applies JEPA pretraining to wireless signals, showing that predicting in latent space builds more transferable representations than pixel-level reconstruction for 5G tasks.
A new framework from USC researchers shows that stacking four distinct knowledge-injection points in diffusion models cuts unsafe outputs by 71%, and no single layer can do it alone.
F3-Tokenizer bridges the gap between audio reconstruction and semantic understanding with a noise-regularized bottleneck and frozen-LLM supervision, enabling a single tokenizer for both tasks.
MResOpt introduces staged residual networks that satisfy constraints in priority order, cutting violations in power grid optimization while staying computationally lean.
GILC steers pretrained discrete diffusion models toward target biological properties at inference time, no retraining required, beating fine-tuned baselines on DNA, protein, and molecule design
A new quantum algorithm samples rare, high-impact events with a provably quadratic speedup over classical methods, without needing to know which events are rare in advance.