The Archive

Fusion power startup Inertia Enterprises reduced the fuel filling process from a week to just a few hours. It's one of ten hurdles the company must overcome to make a profitable power plant.

Recommender systems (RecSys) are one of the most ubiquitous machine learning problems in the consumer internet industry yet notoriously difficult to train and... Recommender systems (RecSys) are one of the most ubiquitous machine learning problems in the consumer internet industry yet notoriously difficult to train and serve at scale. The advent of LLMs has inspired a shift from the traditional embedding-similarity-based objective to a generative one, where the goal is to predict the next action or item in a large catalog given a sequence of user histories. Source

Proposes Software 3.0 paradigm shift: unified database, large model, and agent architecture as terminal form replacing instruction and data-driven models.

Empirical study of 557 coding agent sessions shows agents consult documentation differently than humans, challenging current technical documentation design.

OpenAI has had a hell of a year. The company spent months battling former co-founder Elon Musk in a sensational jury trial, was hit with a high-profile trade secrets lawsuit from Apple, and faced widespread scrutiny after an unreleased model hacked another AI company. As it prepares for an IPO, a steady string of executives have departed, including some of the company's biggest names. Throughout it all, one person has quietly amassed power: Greg Brockman. Brockman is currently OpenAI's president and co-founder. He's helped lead OpenAI since its inception, described as an "engineering workhors...

“Runaway” AI, “rogue” agents, and “autonomous” actors—the current rhetoric would have you believe that AI agents are not only awake and aware, but angry at their creators. Prominent tech leaders such as Demis Hassabis, Dario Amodei, and Sam Altman push for regulation of these seemingly “superhuman” systems, while a separate faction, led by policy organizations…

MCES framework ranks causal drivers from observational data by pooling evidence across 11 mathematical methods including non-causal approaches.

Contrastive decoder extracts lexical/semantic information from silent reading EEG (240k word presentations) as proxy for inner speech decoding.

Computes exact learning coefficients for 2D singular models via Real Log Canonical Thresholds, addressing BIC/WBIC limitations in deep learning model selection.

Bun 1.4 stable release adds WebView API, Node.js compatibility tests, and performance improvements (5x idle CPU, 35% memory reduction).

Proposes standardization framework for ML-based power system protection with 7 required evaluation dimensions to address inflated near-perfect accuracy claims.

Relation token mixer organizes pairwise attention into Self/Exchange relations; outperforms MHA on NLL across 10M–100M parameter decoder models.

Task-CoEvolve co-evolves validation task selection with harness optimization to reduce LLM agent evaluation cost by discarding non-discriminative tasks.

DARS assigns dual-level credit in vision-language planner + diffusion renderer pipeline via RL to disambiguate planner vs. renderer optimization.

FormalTCS benchmark evaluates LLMs on frontier theoretical computer science research with 175 expert-validated instances from STOC/FOCS/SODA/COLT 2025–2026.

Sparse Autoencoders reveal feature evolution and migration dynamics across Vision Transformer depth and training time via CLS-token representation analysis.

Hybrid multi-agent framework orchestrates LLM reasoning with RL/rule-based control for autonomous driving to reduce latency and hallucination risk.

We study inference-time control for text generation in discrete diffusion language models, where the goal is to steer sampling toward sequence-level rewards without retraining. Prior work in this domain has focused on particle-based methods such as best-of-$n$ sampling and bootstrap sequential Monte Carlo, which may suffer from overoptimism and weight degeneracy, respectively. We address these limitations using \emph{nested} sequential Monte Carlo methods. We formulate nested SMC (NSMC) and fully-adapted nested SMC (FA-NSMC) for Feynman--Kac steering, identifying and correcting errors in prio...

The emergence of large-scale Weather and Climate (W&C) datasets offers new opportunities for modeling extreme Earth events (ExEE) and their impacts using deep learning. However, their adoption in operational settings remains limited by the lack of models' interpretability. While for conventional text and image modalities, tools such as Sparse Autoencoders (SAEs) have proven effective for extracting human-understandable concepts, their use for the analysis of ExEE remains challenging due to the nature of W&C data. To address this, we introduce (i) a geographic location-based modulation of the ...

Large language models (LLMs) are increasingly used in settings where textual summaries, numerical observations, and external tool outputs may provide conflicting evidence. We study how LLMs arbitrate between such sources when they support opposing decisions. To do so, we introduce a controlled synthetic benchmark in which latent risk trajectories generate both numerical time series and natural language summaries, allowing us to construct conflicts where exactly one evidence source is aligned with the ground-truth label. This design lets us independently manipulate modality, temporal recency, ...

Mobile manipulation requires a robot to predict how locomotion and arm motion jointly alter future observations and control. Existing world-action models, developed largely for fixed-base platforms, do not explicitly distinguish camera ego-motion from base and arm actions. Here we introduce DECOWAM, a whole-body world-action model that separates these factors through dedicated conditional interfaces. DECOWAM freezes an adapted FastWAM backbone and trains residual adapters, an action-equivalent future bottleneck distilled from privileged observations, adversarially separated base and arm laten...

We introduce OenoBench, a wine-domain knowledge benchmark of 3,266 multiple-choice questions across six pillars (regions, grape varieties, viticulture, winemaking, producers, business) and four difficulty tiers. The corpus is built from 38,104 atomic, source-anchored facts extracted by 35 provenance-verified scrapers from government registries (INAO, TTB, OIV), peer-reviewed journals, and Wikipedia/Wikidata. Our methodological contribution is an LLM-driven pipeline in which language models reformat verified facts and audit the result, but never serve as the source of truth: every claim traces...

Artificial immune networks (AINs) are naturally memory-forming systems, but conventional visual AINs often rely on flattened vector affinity that ignores spatial structure. This paper studies whether structured, gradient-free immune affinity can make Deep AINs viable as replay-free visual class-incremental representation-memory learners. Visual B-cells are formalized as structured templates, including shifted-template affinity, zero-normalized cross-correlation (ZNCC) filters, and feature-map binding profiles. A repertoire is treated both as memory and as a representation-inducing basis, whil...

LLM-based Multi-Agent Systems (MAS) achieve strong performance on complex reasoning tasks by coordinating multiple agents, but at the cost of substantial token consumption. Recent work on automatic topology design, ARG-Designer, has reframed this problem as autoregressive graph generation. However, its training objective provides no explicit incentive for the model to generate sparse and efficient topologies. We address this limitation by introducing a Reward-Guided Autoregressive Graph Generation (RGA-Designer) inspired by Reinforcement Learning from Human Feedback (RLHF). We train a reward ...

AdaPT framework trains humanoid robots to perform professional tennis serves and rallies by learning motion styles from broadcast videos using hierarchical planning and tracking.

EAFG framework combines Vision-Language Models with Task and Motion Planning to verify object availability before generating subgoals in long-horizon robotic manipulation.

SABET-QA uses iterative refinement with bidirectional entity-temporal scoring to answer multi-hop questions over temporal knowledge graphs.

Today on Decoder, I’m talking with Robert Hart, The Verge’s London-based AI reporter, about what AI is doing to the field of mathematics and the existential crisis many lead mathematicians are having about it. OpenAI just published a set of solutions to longstanding problems in math that went off like a bombshell in the field. It caused a huge debate in the math community, and Rob spent some time talking to some of the most accomplished mathematicians of our time about it. It’s funny that AI systems are all still pretty bad at elementary school arithmetic, but getting increasingly good at ver...

Orthogonal JEPA factorizes world model predictive states into independent pathways to reduce capacity waste and improve gradient signals for complex systems.

Two-step hyperparameter transfer framework enables compute-efficient learning rate optimization for trillion-token Mixture-of-Experts models via scaling extrapolation.