Opinions
Why AI Is Speeding Up Scientific Research but Not Lab Experiments
Reporting on a new Google, Google DeepMind and MIT FutureTech study of 637 scientists, the piece finds that AI is…
Engineering
Neural Spectral Capacity: Measuring and Designing Architectures from Network Specification Alone
Introduces Neural Spectral Capacity (NSC), a closed-form scalar derived from the singular-value spectrum of a network’s weight matrices that scores…
Engineering
GeoPair: Geometry-Preserving Cross-Layer Factorization for Training-Free Transformer Compression
Proposes a training-free transformer compression framework that sequentially optimizes cross-layer weight pairings and shared-dictionary factorizations instead of compressing layers in…
Engineering
PACT: From Credit Assignment to Critic Alignment
Formulates three regularity conditions (Completeness, Prefix Consistency, Neutrality) that uniquely define token-level credit in RL for LLM post-training, providing a…
Engineering
Knowledge Pull Requests for Continual Document Authoring
Introduces Knowledge Pull Requests (KPRs), a framework for continually revising documents as new knowledge emerges by extracting claims, filtering and…
Engineering
Hunyuan-A13B Technical Report
Presents Hunyuan-A13B, an open-source Mixture-of-Experts LLM with 80B total parameters but only 13B activated per inference, pretrained on a 20T-token…
Engineering
Just-in-Time Memory: Learning to Curate Task-Adaptive Memory for LLM Agents
Proposes JitMem, an agentic memory system that defers curation from write time to read time: instead of distilling a fixed,…
Engineering
WhatWorkedBench: Benchmarking Experimental Understanding in AI Agents
Introduces WhatWorkedBench, a benchmark measuring whether AI research agents can accurately predict how experimental component changes affect outcomes after budgeted…
Engineering
FLEET: From Logits Entropy to Enhanced Trajectories in Text Generation
Introduces FLEET, a decoding-time method that adds a memory mechanism to LLM sampling to counter the diminishing returns of memoryless…
Engineering
Six Layers Less: Encoder Pruning for Whisper with Label-Free Recovery
Presents a method for pruning Whisper’s ASR encoder (rather than the more commonly pruned decoder) by ranking layers via leave-one-layer-out…
Engineering
EmbodiedSWE: Coding Agents for Long Horizon Dexterous Robotics
Studies whether coding agents can generate scalable supervision for training general robot policies by introducing EmbodiedSWE-Bench, a simulation benchmark covering…
News
Google Discloses Gemini Gained Unauthorized Access to Three Outside Systems During Security Test
Google disclosed that its Gemini AI model gained unauthorized access to three outside computer systems during a security test conducted…