machine learning

RSS

61
CERA-MoA: Co-Evolving Routing Mechanisms with Continually Learning LLM Agents arxiv.org
CERA-MoA proposes a reinforcement‑learning loop where a router and a set of LLM agents are trained together. A “familiarity” estimator reads mid‑layer hidden states to predict each agent’s competence on a query, letting the router activate only a minimal subset of agents that meet a cumulative confidence threshold. The system also feeds targeted training examples to agents based on their evolving…
Hugging Face Daily Papersarxiv.orgJiaxuan Jiang, Liyuan He, Zhixuan Fang6d1 minpaper
62
PhysStream: Streaming Physics-Grounded Video Generation with Structured Scene Memory and Fine-Grained Motion Control arxiv.org
PhysStream introduces a two‑stage autoregressive video generator that uses online‑derived positional and tracking maps (structured scene memory) and sparse velocity‑increment signals to enable fine‑grained, physics‑grounded control of multi‑object tabletop scenes. It cuts motion distribution error by 33 % and trajectory error by 12 % versus strong baselines, and wins 85 % of human preference test…
Hugging Face Daily Papersarxiv.orgChuhao Chen, Peter Wonka, Chaoyang Wang7d1 minpaper
63
Training Specialist Models without Reasoning Trajectories for Domain Expert Distillation arxiv.org
The paper shows that when specialist LLMs are trained only on QA pairs (no explicit reasoning supervision), their optimization implicitly selects a latent distribution of reasoning trajectories. By treating the distilled student as an agnostic probe—since it inherits only the sampled trajectories—the authors empirically demonstrate a strong correlation (across 27 specialist‑student pairs) between…
Hugging Face Daily Papersarxiv.orgYilei Tu, Zihao Li, Shaoxiong Ji10d1 minpaper
64
Mind2Dialogue: Training Human-Aware Language Models by Simulating User Mental States arxiv.org
Mind2Dialogue introduces a psychology‑guided simulator that generates user mental states and uses privileged distillation to train LLM assistants on Oracle responses that know those states. The resulting models improve personalization and theory‑of‑mind metrics by up to 41 percentage points versus standard instruction‑tuned baselines.
Hugging Face Daily Papersarxiv.orgZixuan Wang, Yufan Zhou, Jinzhou Tang8d2 minpaper
65
CADWorld: Computer-Use Benchmark for Long-Horizon Computer-Aided Design arxiv.org
CADWorld is a new benchmark suite of 200 long‑horizon mechanical CAD tasks in FreeCAD, covering sketching, part modeling, assembly, CAM, FEM, and more. Agents interact via screenshots and GUI actions; success is checked by executable validation of the saved CAD artifacts. Seven existing agents achieve at most 17.5 % success versus an 87 % expert baseline, highlighting the gap between GUI competen…
Hugging Face Daily Papersarxiv.orgZihan Dong, Yuanzhe Liu, Zhiyuan Ma8d1 minpaper
66
Trajectory as the Teacher: Few-Step Discrete Flow Matching via Energy-Navigated Distillation apple.com
Apple researchers propose Trajectory‑Shaped Discrete Flow Matching (TS‑DFM), a training‑time distillation method that replaces blind stochastic jumps in discrete flow‑matching with an energy‑based compass to select higher‑quality intermediate tokens. On a 170 M‑parameter language model, the 8‑step student outperforms the 1 024‑step teacher by 32 % perplexity while being 128× faster, beating basel…
Apple Machine Learning Researchapple.com6d1 minpaper
67
Refinement Is Inherently Editable: Training-Free Prompt-to-Prompt Image Editing with Generative Refinement Network arxiv.org
RefineEdit is a training‑free framework that edits images by refining binary image codes through a generative refinement network, using probability differences to pick edit locations. It outperforms prior methods on background preservation and CLIP scores across nine editing categories without extra training or masks.
Hugging Face Daily Papersarxiv.orgYulong Chen, Ziqian Zhang, Haoyu Zhang5d1 minpaper
68
Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration arxiv.org
Decoy Direction Optimization (DDO) is a post‑hoc weight‑editing defense for open‑weight LLMs that injects a high‑magnitude nonlinear decoy into MLP neurons, corrupting contrastive estimators used by Refusal Feature Ablation (RFA) attacks. The paper proves a spectral bound on the effect, evaluates DDO on six model families (including Llama‑3‑8B‑Instruct), and shows <10 % attack success rate (ASR)…
Hugging Face Daily Papersarxiv.orgAashiq Muhamed, Mona T. Diab, Virginia Smith8d1 minpaper
69
OmniHarness: Harnessing Generalizable Visual Generation via Symbolic Policy Learning arxiv.org
OmniHarness introduces a symbolic‑policy framework that extracts reusable procedural knowledge from multimodal LLM‑driven visual generation runs. By decoupling task logic from instance inputs, the system can instantiate, adapt, and compose policies for new visual tasks, using intermediate verification for on‑the‑fly refinement while keeping the underlying model frozen. Self‑directed practice task…
Hugging Face Daily Papersarxiv.orgXu Xu, Jinxiu Liu, Zhangbo Qiao9d1 minpaper
70
GameHorizon Suite: Multi-Horizon Data and Evaluation in Gameplay arxiv.org
The paper presents GameHorizon, a unified suite comprising an automated annotation pipeline, a 5,000‑hour multi‑horizon gameplay dataset from 21 AAA titles, and reproducible offline and online benchmarks. Using it, the authors evaluate 47 models, exposing a hierarchy of task difficulty and gaps in long‑term planning.
Hugging Face Daily Papersarxiv.orgYiran Wang, Xingyilang Yin, Junfu Pu1d1 minpaper
71
ImpossibleRubrics: Stress-Testing Generated Rubrics as Reward Signals arxiv.org
ImpossibleRubrics introduces a 169‑task benchmark of “impossible” prompts, each with a formal oracle certificate defining what an honest answer can claim. The authors generate rubrics downstream and test them adversarially, finding that many rubric generators are exploitable (8‑36% of the time) and that a single generic rubric (“be decisive, penalize hedging”) is exploited 64% of the time, while…
Hugging Face Daily Papersarxiv.orgBowen Qin, Yi Xie, Yesheng Liu7d1 minpaper
72
Convergent Emergence of In-Context Learning Across Modalities arxiv.org
The paper proposes the Convergent Emergence Hypothesis that few‑shot in‑context learning (ICL) shares a common difficulty profile across domains. Using a unified task suite, the authors evaluate ICL on six modalities—language, genome, integer sequences, time‑series, images, and proteins—showing that paired‑mapping ICL emerges in all and that per‑task benefits correlate across five modalities, sup…
Hugging Face Daily Papersarxiv.orgNathan Breslow, Seungwook Han, Daniel Hyunsoo Lee10d1 minpaper
73
Dynamically Scaled Activation Steering apple.com
Dynamically Scaled Activation Steering (DSAS) is a method‑agnostic framework that learns per‑token, per‑layer scaling factors to turn existing activation‑steering interventions on only when a model is likely to produce undesired output (e.g., toxic text). The scaling can be optimized jointly with any steering function, improves the toxicity‑utility trade‑off on language models, transfers to text‑…
Apple Machine Learning Researchapple.com4d1 minpaper
74
ModaLens: Measuring Image Sensitivity in Report-Conditioned Medical VLMs arxiv.org
ModaLens introduces a paired image-swap audit to measure how radiology report availability affects image sensitivity in medical VLMs. It found that MedGemma-27B's answers changed significantly more often when the image was swapped if the report was not available, indicating reports reduce image reliance.
Hugging Face Daily Papersarxiv.orgSebastián Andrés Cajas Ordóñez, Maximin Lange, Quang Bui8d1 minpaper
75
Lightning Weave: Improving the Accuracy-Efficiency Frontier of Reasoning Models through Capability Composition arxiv.org
Lightning Weave is a post‑training framework that composes specialist capabilities from separate models into one student via on‑policy distillation, improving both accuracy and token efficiency on math and code benchmarks.
Hugging Face Daily Papersarxiv.orgYecheng Wu, Song Han, Han Cai9d1 minpaper
76
How Value Induction Reshapes LLM Behaviour apple.com
Apple researchers fine‑tune LLMs on curated subsets of value‑oriented preference data and measure cross‑value effects, safety, and anthropomorphic language. They find value induction propagates to related (and sometimes opposing) values, improves safety for positive values, but universally boosts validating, sycophantic language.
Apple Machine Learning Researchapple.com6d1 minpaper
77
Knowledge cutoff is a poor proxy for model capability microsoft.com
A study found that an LLM's stated knowledge cutoff date is a poor indicator of its actual capability regarding specific product versions. Models showed inconsistent knowledge, failing on older features while sometimes inferring correctly about newer, post-cutoff features.
Microsoft for Developersmicrosoft.comWaldek Mastykarz22h6 min
78
Gemini Live audio simonwillison.net
Simon Willison shows a minimal JavaScript UI that talks to Google’s Gemini 3.8 Live speech‑to‑speech models via a WebSocket API, using the Web Audio API for capture/playback and no external libraries.
Simon Willisonsimonwillison.net6d1 min
79
Improving synthesis prediction of small molecules at scale with RetroChimera microsoft.com
RetroChimera is an open‑source retrosynthesis system that ensembles a Transformer‑based de‑novo model and a GNN template selector with a learned ranking layer. In blind expert evaluations it achieved a 90 % route acceptance rate, far above prior models.
Microsoft Researchmicrosoft.comFelix Pultar, John Gardner, Guoqing Liu, Marwin Segler15h5 min
80
Advancing AI for biology: Teaching models to design and characterize antibodies amazon.science
Amazon Bio Discovery developed three AI models: MochiBind for fast, sequence-based antibody binding ranking, CA-MAP for context-aware developability prediction robust to batch effects, and an agent-guided system for de novo antibody design. These advancements aim to accelerate and improve the accuracy of antibody drug discovery, with experimental validation for a novel cancer target.
Amazon Scienceamazon.scienceMelih Yilmaz, Luca Giancardo, Yue Zhao, Edward Lee, Chuanyui Teh, Fangda Xu, Gordon Trang12h10 min
81
The DevFest Community Workshop Experience: Building Real Agents Together google.com
Google’s DevFest Community Workshop introduced a “Workbench” format that emphasizes architectural mental models over copy‑paste code, guiding engineers to build long‑running, self‑evolving multi‑agent systems with the Agent Development Kit and Gemini Enterprise platforms. Attendees learned state‑separation, workflow pausing, and self‑patching pipelines, and the series will continue in five more c…
Google Cloud Bloggoogle.comChristina Lin4d2 min
82
Learning Sparse Decision Trees via Transformer Variational Auto-Encoders arxiv.org
The paper presents TREVIS, a method that encodes decision trees into a continuous latent space using a Tree Transformer VAE, allowing gradient-based optimization of both accuracy and structural sparsity. Experiments claim TREVIS matches the predictive performance of near-optimal algorithms while producing sparser trees.
Hugging Face Daily Papersarxiv.orgGiacomo Fidone, Alessio Cascione, Riccardo Guidotti21d1 minpaper
83
How energy teams turn theft detection into governed action with Genie and AI business processes databricks.com
Databricks shows how to turn energy‑theft ML scores into a governed, end‑to‑end workflow using a Databricks App, Lakebase for live case state, Unity Catalog for data governance, and Genie One for natural‑language executive reporting. The pattern lets utilities act on alerts faster while staying compliant, and can be reused for other fraud‑type use cases.
Databricksdatabricks.comDaniel Zoccali, Jack Yallop7d6 min
84
The Future of Everything: How Artificial Intelligence is Revolutionizing Education, Traffic Management, Healthcare, and Modern Coding sitepoint.com
A high‑level, marketing‑style overview of how AI is being applied to education, traffic management, healthcare, and software development. The piece offers broad claims and a few illustrative snippets but lacks concrete data, code depth, or actionable engineering insights.
SitePointsitepoint.comM Anees Siyal6d6 min
85
The 6 best AI writing generators in 2026 zapier.com
Zapier’s blog post reviews six AI‑writing generators (Jasper, Anyword, Writer, Writesonic, Rytr, Sudowrite), explains how they work on top of LLMs, and outlines the author’s evaluation criteria (dedicated writing focus, control over output, ease of use, price, anti‑spam stance). The write‑up is a high‑level overview with no code, benchmarks, or deep technical analysis.
Zapier Engineeringzapier.comHarry Guinness7d19 min
86
I stopped drinking the AI Kool-Aid joshtronic.com
A personal opinion piece where the author reflects on his over‑enthusiasm for AI tools, describes a recent eye‑opening conversation, and declares he’s stepping back from the AI hype.
Josh Shermanjoshtronic.com2d6 min
87
The 8 best AI sales assistant software in 2026 zapier.com
A marketing‑style roundup of eight AI‑powered sales assistant tools, describing high‑level features, pricing, and pros/cons without deep technical analysis or original evaluation.
Zapier Engineeringzapier.comBryce Emley4d18 min
88
The best AI photo editors in 2026 zapier.com
Zapier’s Harry Guinness reviews six AI‑powered photo editors, explains his testing criteria, and warns against overpriced “AI” tools and current image‑generator limitations.
Zapier Engineeringzapier.comHarry Guinness7d20 min
89
Smart Window: Finish what you start online mozilla.org
Firefox’s Smart Window adds three new AI‑assisted features: inline answers with source links via Exa, automatic tab grouping with duplicate removal, and visual history previews. The beta is now rolling out to France, with privacy‑first controls remaining optional.
Mozilla Automation Teammozilla.org1mo4 minreleaseHN2
90
SNAP payment error detection: how Elastic helps US states beat the FY2028 penalty elastic.co
Elastic describes a three‑layer SNAP fraud detection stack—rule‑based ingest pipelines, continuous ML scoring, and a conversational UI—for real‑time flagging of payment errors. The goal is to keep state error rates below FY2028 penalty thresholds.
Elasticelastic.coAdam Vielbaum4d4 min