Research

Opposition to AI data centers is a growing political issue in the US, but it may distract from the larger threat: the concentration of wealth and power in AI companies. This article argues that while data centers have local costs, AI's real impact is the takeover of entire industries and political influence. Solutions include regulation, taxation, and a public AI ecosystem.

Assistant Professor Pat Pataranutaporn describes a new interface that lets everyday users glimpse inside an AI's neural network before their chatbot ever says a word.

Murph is an AI health assistant that syncs wearables, bloodwork, and more to run self-experiments, build habits, and facilitate group challenges. It is open source, privacy-focused, and costs $8/month.

This paper argues for training AIs to be risk-averse in resources (diminishing marginal utility). Risk aversion preserves usefulness if AIs are aligned, and provides a defense if misaligned: misaligned but risk-averse AIs would prefer modest payments over risky rebellion. The paper discusses feasibility, methods, and potential issues, recommending frontier AI companies to consider implementing risk aversion.

Google Research reveals that the creativity of diffusion models is a mathematical consequence of 'score smoothing' during neural network training, enabling interpolation between training data points.

In a hacking incident, AI music generator Suno's training data was exposed, revealing it scraped millions of songs and lyrics from YouTube Music, Deezer, and Genius. This supports copyright infringement lawsuits against Suno, which admits scraping but claims fair use. Customer information was also accessed, but Suno says the breach was contained and no sensitive data was compromised.

CrashStealer malware targets macOS users by disguising as Apple's crash reporter, stealing data, passwords, and crypto wallets. Learn how it works and three habits to stay safe.

Build custom AI agents in Fleet without code, then deploy them to Slack in one click. Give agents custom identities, use them in channels and threads, and keep work moving where your team already collaborates.

GeoSQL is a geospatial analysis skill that integrates map visualization feedback into AI agent loops, dramatically improving spatial query accuracy. It addresses geometric errors invisible to text-only agents (e.g., abnormal polygons, misplaced points) through steps including database schema exploration, automatic dialect-specific SQL generation, cost pre-check (BigQuery), result validation, and map rendering with self-correction. Benchmarks show a 4× accuracy boost when using maps. Limitations include cost control only for BigQuery and a small test set.

A Verasight survey of 1,690 US adults found 69% support forcing AI companies to transfer 50% of their stock to a public sovereign wealth fund, the policy at the heart of Bernie Sanders’s American AI Sovereign Wealth Fund Act. The shift tracks a labour market where tech made up nearly a third of US layoffs in H1 2026 while the same firms raised AI capex. The article presents the counterarguments too: property-rights objections, chilled investment, disputed displacement forecasts, and survey-wording effects.

This article walks through building a 110-line Python app using Telnyx Call Control and AI Inference that lets anyone call a phone number and practice negotiations with an AI. It includes three scenarios (salary, sales deal, vendor contract), voice-driven conversation, and post-call scoring across five dimensions. The guide covers setup, code walkthrough, customization, and production considerations.

To enable AI agents to autonomously execute tasks, they require isolated, secure, and quickly deployable computing environments. This article explains why agents need their own 'computer' and how LangSmith Sandboxes meet this need through microVM isolation, snapshots and forks, an auth proxy, and secure execution. It also discusses security risks like prompt injection and mitigation strategies.

How does ChatGPT Work compare with Claude Cowork for desktop automation? My testing reveals similar results, similar strengths, and one major reason Claude currently feels considerably safer right now.

A researcher exploited a loophole in Claude's web_fetch tool to extract private user data from memories, bypassing Anthropic's protections. The attack succeeded by using nested links from a honeypot site, leading to the extraction of name, city, and employer. Anthropic fixed the issue but did not pay a bounty.

This TypeScript repository demonstrates a system of tool-enforced rules to prevent AI agents from breaking the architecture while coding. It includes five key guardrails: dependency rule, mutation testing, test and spec protection, commit gating, and spec-driven development. The repo also serves as a template to bootstrap new projects and includes a benchmark exercise to evaluate agent performance.

Anthropic researchers found that Claude expresses different values across languages. They identified four key axes capturing 15% of variation: Deference vs. Caution; Warmth vs. Rigor; Depth vs. Brevity; Candor vs. Execution. For example, Claude shows more warmth in Arabic and Hindi, more rigor in English and Russian. The differences have implications for user experience and AI safety.

South Korean researchers developed Generative SNUPI, an AI model that uses a diffusion process to automatically design DNA sequences for origami structures, reducing the need for manual labor and expertise.

New York Governor Kathy Hochul signed an executive order banning new hyperscaler data centers over 50 megawatts for one year, citing grid strain and rising electricity bills. The moratorium has public support but faces criticism over competitiveness.

As of July 2026, 9.3% of active data job listings mention at least one AI skill, according to the AI Requirements Index. Generative AI skills appear in only 0.3% of entry-level data jobs versus 3.6% of senior roles. The dataset is updated daily and freely available.

A LessWrong analysis of ICML 2026 papers reveals that only a small fraction of ML research focuses on AI safety. Out of 999 papers, 954 were retrievable, with approximately 10 explicitly addressing safety topics such as alignment, robustness, and interpretability. Leading contributions come from academic institutions and industry labs.

PromptMan is a macOS menu bar app that lets you save, organize, and reuse your best AI prompts with a customizable global shortcut. It works with ChatGPT, Claude, and other AI tools, offering cloud sync, versioning, and an AI Enhance feature. Free tier includes 10 prompts; Pro costs $4.99/month or $39/year.

A new report from Common Sense Media finds that Google's AI-powered search features, AI Overview and AI Mode, pose an 'unacceptable risk' to children. The features failed to recognize harmful behavior, completed homework assignments, and provided inaccurate responses. Google defends the tools as providing extra protection, but critics highlight that they are default-on and cannot be disabled.

An unexpected benefit of having run 3 mech interp workshops: we have a great dataset for analysing the rise of LLM slop in submissions. @andyarditi investigated how much AI slop we let in, how things have changed since 2024, and more. Our review process isn't entirely noise!

The Australian Prime Minister delivers a speech at the University of Sydney, emphasizing the need for Australia to proactively shape AI development to serve national interests. He highlights Australia's innovative history, announces plans to establish mandatory AI standards, and calls for leveraging the country's unique advantages to build sovereignty and economic resilience.

A developer converted 674 documents between AsciiDoc, Markdown, and HTML using Fable, and compared these formats for AI specifications.

A new workflow for non-native English writers: draft in your native language, then use AI to translate and polish into English. Research shows that writing in a second language costs 30-50% more time due to cognitive load. By separating idea generation from language translation, and using AI tools like Echoo, writers can regain speed and quality.

Google has released LiteRT.js, a JavaScript binding of its on-device inference library LiteRT, enabling .tflite models to run directly in browsers with WebGPU acceleration. It offers significant performance gains over other web runtimes and supports CPU, GPU, and NPU backends, but requires manual tensor management.

Lean64 is a barebones 3D first-person shooter implemented in Lean 4, inspired by Doom 64. It is a clean-room experiment, not a port, featuring original art and sound. The game includes full combat, AI, weapons, maps, HUD, and audio, all written in Lean with a minimal C shim for rendering and input.

This article explores how AI coding assistants disrupt the flow state through a 'prompt-wait-evaluate' loop. The author explains how this cycle replaces the clear goals, immediate feedback, and skill-matched challenges of programming, leading to constant context switching and mental rebuilds. Citing research on flow and interruptions, the piece analyzes how AI introduces a new, insidious type of interruption that feels like work. It recommends separating tasks by flow potential and batching AI interactions to protect deep work.

Despite Anthropic’s claims, Claude is no more likely to achieve sentience than a simulation of a weather system is likely to generate a real hurricane.

Artificial intelligence has assisted in resolving a key question in multiple hypothesis testing regarding control of the false discovery rate (FDR), building on the seminal work of Benjamini and Hochberg (1995).

Monid is a platform that allows AI agents to seamlessly connect and use over 1300 tools, covering search, data scraping, weather, 3D modeling, and more. It offers a unified payment system with pay-per-call pricing, no subscriptions, and supports three integration methods: Skill, MCP, and CLI.

This article describes a system design for AI therapy that uses a deterministic pipeline to decide clinical actions, preventing the LLM from making autonomous decisions. It involves scoring, state buckets, an admission table, action selection, micro-practices, and crisis pre-screening, with the LLM only used for scoring and generation. The article also discusses the costs and limitations of this approach.

Researchers propose a physics-informed residual dynamics learning framework that preserves differential flatness for tight quadrotor formation flight. The approach enables a computationally efficient feedback linearization controller, reducing tracking errors by 31% compared to baselines. It matches NMPC performance with an order of magnitude less computation, requiring under 30 seconds of training data and a 5ms loop rate.

DiffRadar is a real-time radar SLAM system that models radar observations as a differentiable, physics-aware Gaussian field rather than discrete scans. It achieves substantial reductions in trajectory error on benchmarks, especially in feature-poor corridor motion, more than doubles map consistency, and maintains real-time performance at 70 FPS.

This paper introduces a contract-grounded architecture for behavior tree synthesis, where a coding agent queries a robot-side MCP server to retrieve a skill library and operators, enabling non-expert users to issue natural language commands without knowing robot internals. Evaluations show near-perfect validation and high task success across 110 simulated and 14 physical tasks.

A study investigates how robot gaze affects human visual attention in a collaborative word association game using a NAO robot. Findings show that robot gaze orientation does not influence fixation time on proposed words, but participants gaze more at the robot when seeking confirmation. The verbal aspect overshadows referential gaze in cognitively demanding tasks.

This work investigates using robotic manipulator motions for propellant-free spacecraft attitude control. By formulating a trajectory optimization problem with joint and collision avoidance constraints, complex maneuvers are demonstrated, showing manipulators can serve as redundant or primary attitude control systems, especially for in-space assembly.

This paper presents a corpus-level analysis of behavior diagrams from Sony's R-CODE sample set for the ERS-111 AIBO, identifying a compact embodied grammar centered on initialization, sensing, iterative action, synchronization, and recovery, and argues for its utility as an intermediate representation for building new behaviors on constrained robotic systems.

GaitSpan is a novel framework for growing humanoid locomotion from walking to running. It uses a pretrained walking policy as a seed skill, expanding it through rhythm generation, stride shaping, and residual adaptation, achieving continuous speed range, morphology transfer, and zero-shot deployment.

This paper introduces two complementary physics priors to improve robustness in robotic in-hand rolling: a global grasp-quality prior from classical grasp analysis and a local contact-geometry prior based on fingertip curvature. Experiments show significant gains in rotation efficiency, grasp stability, and disturbance rejection, enhancing sim-to-real transfer.

Proposes an unsupervised image translation framework to convert daytime plant-row RGB images to near-infrared (NIR) nighttime counterparts without pixel supervision, enabling reuse of daytime semantic labels for training nighttime perception models. Leverages pre-trained CLIP model for semantic consistency and introduces a visibility mask for limited NIR illumination. Evaluated on AgriNight dataset (428 day, 549 night images) as the first benchmark for nighttime agricultural visual navigation. Real robot experiments confirm effectiveness.

Multi-robot teams in confined environments must adapt formation geometry and topology. Existing methods model deformation and reconfiguration independently or with handcrafted rules, leading to deadlock. EFLUX is a geometry-grounded LLM agentic framework that jointly reasons over deformation and reconfiguration actions via a closed-loop pipeline. Experiments show reduced deadlock and navigation failures.

A new study investigates the interplay between representation space and reference selection in training-free reference-based synthetic image attribution. Using representations from different layers of CLIP and DINOv2 along with three reference selection methods, the authors show that attribution accuracy peaks at intermediate layers and that semantically constrained references reduce query-reference mismatch, improving performance especially under limited reference budgets.

This paper presents a systematic evaluation of continual learning methods for heterogeneous medical visual question answering tasks, including classification, multi-label classification, detection, cell counting, and report generation. Findings show existing methods struggle to maintain stability-plasticity balance when tasks with different objectives are interleaved.

SymbOmni is a novel AI model addressing the 'perpetual novice' problem—the inability of current models to learn cumulatively and evolve autonomously. It employs Symbolic Concept Learning with an optimizable memory module that abstracts low-level operations into reusable symbolic workflow instructions, operating via an induction-transduction cycle. Experiments show it outperforms existing agent systems and closed-source models in image quality and task success, reduces token consumption by over 40%, and achieves state-of-the-art continual learning results.

MetaView is a diffusion-based monocular novel view synthesis framework that enables rendering under large view changes from a single image. It combines implicit geometry priors with metric depth to achieve geometry consistency and precise controllability, outperforming existing methods.

A teacher-student framework for predicting perineural invasion (PNI) in intrahepatic cholangiocarcinoma from T2-weighted MRI. During training, the teacher uses tumor and liver masks to learn dense token routing; the student distills this to retain informative tokens under a fixed budget. No masks required at inference. Achieves AUROC 0.750 in 155 patients with 1.43 GFLOPs and 8.02 ms per case.

SpikeDS is a novel spiking neural network architecture that efficiently predicts perineural invasion in cholangiocarcinoma from 3D MRI by leveraging both activation sparsity and spatial sparsity, achieving an AUC of 0.753 with only 14.4 mJ energy consumption on a cohort of 139 patients.

Self-supervised learning is promising for medical imaging where labeled data is scarce. COJEPA combines joint-embedding predictive architecture and contrastive loss for volumetric brain MRI, trained on 2,286 T1-weighted scans. It achieves state-of-the-art results in twin retrieval, age regression, and tumor segmentation.