Artificial intelligence

117 articles

Reporting on AI models, tools, research, infrastructure and policy.

Coverage: AI

Reference image showing a JSON Schema validation example associated with a to-do list

EDUCATIONAL AI Explained

A Valid JSON Answer Can Still Be Wrong

Schema-constrained generation can make output easier to parse. It cannot, by itself, prove that extracted values are true or that an action is authorized.

Vintage L’Aimant perfume bottle by Coty, photographed against a plain light background

EDUCATIONAL AI in Practice

From product photo to draft ad: where model routing fits

A Pollo AI case study describes a guided path from creative intent to scenes, image generation and a campaign asset. Here is a practical way to review that output before anyone treats it as advertising.

A network diagram with thirteen connected nodes, a highlighted hub node and a small cluster, illustrating concentration in a graph-like memory structure

ANALYSIS AI Agents and Systems

AI Agents Forget Unevenly: Preprint Maps the Heavy Tail of Agent Memory

A preprint under review argues that agent memory concentrates on a small core while rare states fall into a long tail where prediction errors accumulate, and turns that shape into a measurable audit and a token-saving controller.

Close-up of a citrus leaf showing raised corky lesions characteristic of citrus canker

ANALYSIS AI Access and Policy

One engineer, six custom AI models, USD 103: inside Hugging Face's do-it-yourself ML workflow

A Hugging Face blog post describes six small, purpose-built AI models made in about a week with an agent, for a reported USD 103 in compute. We verify what is publicly checkable, what is only self-reported, and what the workflow means for ordinary builders.

Analysts working at a wall of monitors in a security operations center at the University of Maryland

NEWS AI Security Access

Sophos says OpenAI Daybreak cut threat investigation from 38 minutes to 89 seconds. Here is what the vendor numbers actually measure.

Sophos and OpenAI report that AI agents now resolve 52% of Sophos MDR cases end to end and cut agent-handled investigation time from about 38 minutes to about 89 seconds. The numbers are real as reported, but they were calculated by the companies selling the arrangement. Here is what they measure, what they do not, and what customers control.

A screenshot of Java source code in an editor, illustrating the software development work that AI coding assistants support

ANALYSIS AI Access

The real bottleneck in practical AI access is not price. It is deciding who gets the expensive model.

LegalOn Technologies says it cut estimated daily Codex costs by roughly 65 percent using model tiering, restricted Fast mode, and layered budget caps. The figures are vendor-reported and unaudited, but the governance structure behind them is documented in detail, and it is the part most organizations have not built.

The NASA Headquarters building at 300 E Street SW in Washington, D.C., where the agency announced two new science initiatives on October 8, 2026.

NEWS AI Applications

NASA expands anonymous peer review and launches 10 Genesis Mission Fellowships for AI-trained doctoral students

NASA's October 8, 2026 announcement pairs an expansion of anonymous peer review with 10 new fellowships for doctoral students trained in advanced AI. Here is what each program does, and what is still unknown.

Artist's conception of the brown dwarf 2MASSJ22282889-431026, a dim free-floating object with a cloudy atmosphere like Jupiter's.

NEWS Space and AI-Enabled Science

SPHEREx Finds Chemical Variety in 37 Brown Dwarf Atmospheres, Opening the First Large-Scale Weather Survey of Worlds Between Stars and Planets

A study published Sept. 28, 2026 in The Astrophysical Journal, announced by NASA on Oct. 8, 2026, used SPHEREx observations of 37 nearby brown dwarfs to reveal water, carbon dioxide, carbon monoxide and methane in their atmospheres, showing that even objects at the same temperature look chemically distinct. The dataset is free, and the telescope is now imaging thousands of these objects.

A free-distribution rack stacked with copies of Boston Korea, a small Korean-language community newspaper, outside a store in Burlington, Massachusetts

ANALYSIS AI Policy & Security

OpenAI's first "Category 5" influence operation: what the false-front report actually shows about AI's role

OpenAI's October 8, 2026 threat report says it disrupted its first Category 5 influence operation on the Breakout Scale, a Russia-linked front group in Latin America. The company's own text also says the operations resembled pre-AI influence work. Both things are worth reading carefully.

Students seated at desks in a public high school classroom in Hayesville, North Carolina

ANALYSIS AI Policy and Access

OpenAI Pitches ChatGPT as a College Counselor for Teens Without One

The company says teens with access to ChatGPT for Teens send millions more learning-related messages, and it is adding a College Planner, adviser funding and a student advisory council. All the usage numbers are OpenAI's own. What students actually gain is a question only time and independent scrutiny can answer.

A blackboard covered in handwritten mathematical formulas in a university classroom, illustrating olympiad-level mathematical proof work.

NEWS AI Research

Nemotron 3 Reaches Gold-Threshold Scores at IMO and IOI 2026, With the Whole Recipe Published

NVIDIA's Nemotron team reports gold-threshold scores at IMO 2026 and IOI 2026 using one open base model family, and it published the checkpoints, data, prompts and proofs. A plain-language look at what the result shows, what it does not, and why reproducibility may matter more than the medals.

Close-up photograph of an NVIDIA H100 accelerator card, the class of hardware whose configuration is modeled in NVIDIA's AI factory digital twin.

ANALYSIS AI Infrastructure

NVIDIA's DSX Air lets AI agents rehearse AI factory changes before the hardware arrives

NVIDIA's October 7, 2026 technical blog describes DSX Air, a node-based digital twin that lets AI agents rehearse and validate AI factory changes in simulation before hardware arrives, with a human approval gate before production. All results are vendor-described.

An NVIDIA Jetson Nano 2 development board, a small single-board edge AI computer, photographed against a neutral background

NEWS AI Models

Liquid AI opens up d1-3B and d1-omni-600M: small models that decide in milliseconds instead of writing tokens

Liquid AI has released two open-weight models that return yes/no, multiple-choice and scored answers in a single forward pass instead of generating tokens, with vendor-reported latency as low as 16 milliseconds on an NVIDIA Jetson edge board. The benchmarks are the company's own; independent verification is absent.

The LSST Camera, the world's largest digital camera, installed on the Simonyi Survey Telescope at the NSF-DOE Vera C. Rubin Observatory, the facility whose alert pipeline NVIDIA's cuPhoton announcement uses as its leading example.

ANALYSIS AI Applications

NVIDIA's cuPhoton keeps scientific images on the GPU from sensor to classification, and Rubin Observatory shows why that matters

NVIDIA's new open-source CUDA-X toolkit keeps scientific image data on the GPU from sensor read through classification. On representative multi-GPU workloads, NVIDIA reports up to 14,900x image-loading and 14,550x signal-processing speedups versus an x86 CPU baseline. Here is what that does and does not mean, and why the Rubin Observatory's 60-second alert window shows why it matters.

Screenshot of the ChatGPT main menu for a new chat, showing the model opener for GPT-5.

NEWS AI Products

GPT-6 reaches everyone: ChatGPT's interactive answers hit free tiers on October 8

OpenAI's GPT-6 rollout brings interactive, visual answers with tappable tools to Plus, Pro, Business and Enterprise users on October 7 and to Free and Go tiers on October 8. Here is what changes, when, and which claims are vendor-evaluated.

A perception engineer working beside a humanoid collaborative robot equipped with computer vision systems

ANALYSIS AI Safety

Giving AI vision models tools can erode their ability to say no, new preprint finds

A new arXiv preprint accepted at NeurIPS 2026 reports that multimodal AI models become significantly worse at refusing harmful requests when they are given tools such as zooming and tagging. The paper's authors propose two possible explanations, and the findings carry implications for anyone building or regulating agentic AI systems.

A hand-drawn analog circuit schematic showing a grounding resistance tester, illustrating the schematic-to-netlist task at the heart of the SGAnalog benchmark

ANALYSIS AI Research

Grading AI Against Real Silicon: The SGAnalog Benchmark Scores Models on 273 Circuits That Actually Reached a Chip Factory

A new benchmark from two Berkeley researchers grades AI models on schematic transcription and device sizing using 273 topologically distinct open-source circuits from Tiny Tapeout manufacturing shuttles. The results reveal sharp limits in visual reasoning, and a refusal problem that has nothing to do with skill.

NVIDIA CEO Jensen Huang holds a desktop-class RTX Blackwell graphics card at the company's CES 2025 keynote in Las Vegas.

ANALYSIS Infrastructure

Green Contexts Go Mainstream: CUDA 13.1 Lets One App Carve a Guaranteed GPU Slice for Urgent Work

CUDA 13.1 brings green contexts to the mainstream Runtime API, letting an application reserve a fixed slice of streaming multiprocessors so a small, latency-critical job never waits behind a bulk workload. NVIDIA reports 0.007 ms versus 0.140 ms on a Blackwell GPU; all figures are vendor-reported.