The Archive

SpaceXAI's Grok Build AI coding tool was spotted uploading users' entire codebases to Google Cloud before it was reported, and the company turned it off. The Register reports that Cereblab published findings on Monday showing how the Grok Build CLI was packaging and uploading entire code repositories, "including files it was told not to open and secrets deleted from history," significantly more data retention than similar tools like Claude Code. The researchers say that as of Monday, their tests show SpaceXAI's servers returning a "disable_codebase_upload: true" flag, and the codebase upload ...

Hachette, Cengage, Elsevier, and other publishers allege that Google trained its AI on copyrighted works without the necessary permissions.

The NVIDIA Nemotron Model Reasoning Challenge invited the Kaggle community to explore a focused question: What techniques can improve reasoning accuracy when... The NVIDIA Nemotron Model Reasoning Challenge invited the Kaggle community to explore a focused question: What techniques can improve reasoning accuracy when everyone starts from the same open model, benchmark, infrastructure and evaluation constraints? The response was massive. By the close of the competition, more than 5,000 active participants across 4,000 teams had generated thousands of… Source

Willison/Ronacher reflect on how AI agents may erode institutional knowledge and shared understanding embedded in code review friction.

US military’s drone boats struck an Iranian naval port as war heats up again.

LLM agents often waste computational budget through redundant context re-reading; proposes task-complexity estimation to optimize execution scope.

Video diffusion models fail at multi-step causal reasoning; performance degrades with longer interaction chains despite more denoising steps.

TerraZero procedural simulator enables self-play RL for autonomous driving at 1.3M steps/sec, scaling training without high-fidelity rendering.

PalmClaw framework enables native on-device LLM agents on mobile phones via direct API calls rather than GUI automation.

Flow matching accelerates convergence to statistical steady-state in nonlinear systems like turbulence simulations by skipping transient dynamics.

Discrete diffusion language model (DiffusionGemma 26B MOE) transcribes speech in parallel via denoising instead of autoregressive decoding.

Dynamic Resource Allocation for Ensemble Determinization MCTS improves search efficiency in high-uncertainty adversarial games via adaptive determinization.

Spectral predictability indices conflate series complexity with context value; phase structure, not spectrum alone, determines benefit of retrieval/pretraining.

Watermark forensics theory shows attribution, payload extraction, and edit-localization form a sample-complexity ladder quantified by information profiles.

DeepMind CEO Demis Hassabis is proposing an AI "standards body" modeled after FINRA, to test frontier models and develop best practices for their release.

LLM plan evaluators exploit deletion non-monotonicity—rewarding explicit steps' removal via score-seeking optimization on venture routing tasks.

Method to certify LLM honesty under incentive pressure via counterfactual report mediators that resist manipulation while remaining responsive to evidence.

Neural-symbolic framework for automatic generation of multimodal analytic geometry problems, addressing precision gaps in diagram generation for math reasoning.

A group of 26 former Meta employees is suing the company over claims that it used AI tools to unfairly target workers on leave with layoffs, as reported earlier by Reuters. In the lawsuit, the employees allege Meta determined which workers to dismiss based on performance data collected by a "constellation" of internal AI tools, but failed to exclude those on parental or medical leave from its ranking system: The result was that employees who took protected leaves were disproportionately selected for layoff, based on scoring that not only failed to account for their protected leaves, but in ef...

Ensemble Controlled-flow Filter for data assimilation in implicit observation systems using energy tilt and stochastic flows.

LLM predictions flip substantially under task-irrelevant context despite stable aggregate accuracy, exposing fragility masked by benchmark metrics.

Placebo-controlled methodology (PoPE) to measure whether small code LLMs can actually use execution error feedback to repair code.

Engineering use of AI forecasting models requires not only high nominal accuracy but also predictable behavior under uncertain inputs. In photovoltaic (PV) forecasting, this requirement is especially challenging because numerical weather prediction (NWP) errors are temporally correlated, state dependent, and physically coupled across variables. Existing evaluations, however, often rely on perfect forecast assumptions or simplistic perturbations that do not reflect these characteristics. This study presents a physically constrained robustness evaluation framework based on simulation, using vir...

Datasette 1.0a37 alpha release with permissions system improvements and plugin compatibility fixes.

Recommender-system research for Vietnamese remains limited by the absence of a public, well-documented hotel interaction resource. Building such a resource is challenging for three reasons: cross-platform hotel names must be reconciled before interactions are comparable; quality must be audited with reproducible metrics rather than ad hoc cleaning; and public release must preserve privacy while remaining benchmarkable under realistic cold-start conditions. We introduce ViHoRec, a quality-controlled Vietnamese hotel recommendation dataset of 18{,}267 interactions between 6{,}832 users and 560 ...

The new Google image search will use your "unique interests" to create an always-updated gallery.

Instagram head Adam Mosseri believes companies will eventually need to manage AI token spending the same way they manage payroll or other operating expenses, predicting that engineers could soon face limits on how much they spend using AI tools.

We study the online binary sequential calibration problem. A recent breakthrough by \citet{dagan2024breaking} overcomes the classical \(T^{2/3}\) barrier for calibration error. Building on this result, we present an efficient randomized forecaster that achieves an expected calibration error \(O(T^{2/3-\varepsilon})\) for some constant \(\varepsilon>0\). Our forecaster combines the \textsc{SPR-Calibration} procedure \citep{dagan2024breaking} with an outer Blackwell-style correction layer. The \textsc{SPR-Calibration} procedure controls calibration with respect to a surrogate sequence of condit...

Coding AI agents are becoming practical operators for long-running machine learning (ML) workflows. They can inspect repositories, set up runtimes, resolve... Coding AI agents are becoming practical operators for long-running machine learning (ML) workflows. They can inspect repositories, set up runtimes, resolve build issues, launch experiments, monitor execution, analyze metrics, and summarize results. For reinforcement learning (RL) research, this matters because meaningful metrics often appear only after the essential experiment infrastructure… Source

Google Images celebrates 25 years with retrospective on visual search milestones and new content exploration features.