Do Multimodal RAG Models Ignore Late Evidence? A Primacy Bias Test
Multimodal RAG readers lose 16 to 26 percentage points when the correct evidence sits at the end of context, and standard rerankers do not close the gap.
The archive · Page 4 of 6
Models & Research
Where architecture, training tricks, and eval methodology meet the marketing layer — separating durable progress in foundation models from leaderboard theater that quietly falls apart under load.
The archive
73–96 of 139 articles · Newest first
Can Deep Learning Design RF Power Amplifiers Without Full EM Simulation?
A Chalmers/Tampere paper trains a CNN on EM-simulated layouts to search Doherty amplifier combiners in milliseconds. EM simulation is amortized, not eliminated.
Synthetic Clinical Notes from LLMs: Believable Prose Is Not Clinical Validity
A 70-patient synthetic EHR pipeline shows believable prose is not clinical validity. Fabricated dosages and labs poison models and invite HIPAA and FDA scrutiny.
Doubao vs Qwen 3.7 vs GLM-5.2: Route by Axis, Not Leaderboard
Doubao-Seed-2.1 Pro's parity claims are vendor-graded and unreproduced. Route the Chinese flagship tier by axis: Qwen for ZH to EN, Doubao for EN to ZH, DeepSeek for code.
Can Dynamic Experts Fix Catastrophic Forgetting in Robot Manipulation?
LiMoDE freezes a pre-trained expert bank and adds low-rank experts per new task, forcing continual robot learning to choose between architecture and replay.
Error-Conditioned Neural Solvers vs Iterative Refinement: When Does Learned Correction Win?
A June 2026 preprint feeds the PDE residual into a neural corrector as input, not an optimization target, shifting surrogate cost from inference loops to training capacity.
Vision-Language Models Move Past Object Detection: The MLLM Perception Shift
Vision-language models now reason over tables, charts, and documents, but detection-era benchmarks still rank them on box localization and undercount comprehension.
Can Autoregressive Boltzmann Generators Replace MCMC in Simulation?
ArBG reframes equilibrium sampling as one forward pass and beats flow-based generators, but MD training data and importance-sampling reweighting remain in the pipeline.
Look-Before-Move Plans Observation Before Motion in Dynamic 3D Story Worlds
The June 2026 preprint Look-Before-Move separates what a camera observes from how it moves in dynamic 3D story worlds, but its gains stay preprint-only with no numbers.
GLM 5.2, Qwen 3.7, and DeepSeek in 2026: A Routing Map by Workload, Not by Rank
GLM 5.2 sits within 1% of Opus 4.8 on FrontierSWE, Qwen 3.7 Max owns agentic workflows, and DeepSeek V3.2 covers cheap bulk inference. Route by workload, not by rank.
SLM Pipeline Catches 10% of Papers Human Reviewers Missed, but No Model Matched Human Accuracy
An SLM ensemble caught 10% of papers human reviewers missed, yet no model matched human accuracy. Now journals must define how much SLM-screened evidence they will accept.
MiniMax M3 vs GLM-5.2: Whose 1M-Context Claim Holds Up?
Both MiniMax M3 and GLM-5.2 advertise 1M context. Only Z.ai ships benchmarks, and neither publishes independent needle-in-haystack data, so the spec sheet is not the budget.
DeepSeek V4.1 Flash vs Qwen 3.7 vs Llama 4.5: June 2026 HF Trending Ranks Velocity, Not Installs
DeepSeek V4.1 Flash led Hugging Face trending in June 2026 within one week, and five of the top ten slots went to Chinese labs. Trending measures velocity, not installs.
125 Targeted Wikipedia Edits Left a Detectable Signal in Llama Pretraining
125 Wikipedia edits produced significant attribution signals in Llama 3.1 8B and Llama-3.2-1B, showing pretraining does not average out targeted high-weight-source edits.
Can SAE Features Stop LLMs From Forgetting During Continual Learning?
A June 2026 preprint anchors continual-learning retention in sparse autoencoder features instead of weights, removing replay buffers but tying quality to SAE coverage.
Can a 30B Model Post-Train Itself? A-Evolve-Training Tests Autonomous RL
A 30B Nemotron model post-trained itself to 8th of 4,000 on NVIDIA's leaderboard, then detected its own internal metric lying and rewrote its evaluation frame mid-run.
Compare DeepSeek, Qwen and GLM leaderboards by artifact and task
A leaderboard rank is only comparable when the models, revision, precision, prompt protocol, and benchmark version match.
Qwen3.7-Max's Top-Ranked Claim vs the Artificial Analysis Index
Alibaba framed Qwen3.7-Max as globally top-ranked at launch. The Artificial Analysis index scored it 56.6 and ranked it fifth overall, behind Claude and GPT.
Does Tree-of-Thought Reasoning Scale to Billion-User Modeling?
ScaleToT distills tree-of-thought reasoning into a profile encoder that serves billion-user recommenders without per-user LLM calls, lifting LT30 by 6.738% in an A/B test.
Can LLMs Debug Verilog? VeriPilot Puts an Agent on RTL Errors
VeriPilot wraps an LLM in a four-phase Verilog debug loop with a golden reference, lifting GPT-4o repair on CVDP from 54.3% to 85.71% in an unreproduced v1 preprint.
Task Decomposition Helps LLMs by Shrinking Output Space, Not by Cutting Labeling Cost
No preprint backs decomposed annotation as an LLM labeling-cost fix. Where measured, gains come from shrinking a model's output space, not from splitting annotation work.
Flow Matching vs U-Net: A Skip-Free Backbone for Speech Models
A June 2026 preprint argues U-Net skip connections leak noise into flow-matching speech decoders, and a codec-supervised skip-free backbone can replace them at parity.
A Per-Neuron Sequence Model Was Withdrawn From arXiv as Coverage Hailed It
TND proposed per-neuron dynamics as a sequence-modeling primitive, then was withdrawn from arXiv for accuracy errors the day before coverage called it a Transformer.
PV-TAM Corrects Decoding Drift and Boundary-Marker Bias in VLM Localization Scoring
PV-TAM, proposed in a June 2026 arXiv preprint, corrects two structural biases in VLM localization scoring: decoding drift and modality boundary-marker contamination.
