Explainers
Longform AI explainers with context, source trails, and related stories.
Showing the latest 40 of 449 published records. Detail pages remain available through sitemap and internal links; date archives are the next production step.
A new framework allows for changing the environment and lighting in videos while preserving the character's facial expressions and features based on a single keyframe.
As part of the CHIPS Act, the US government is awarding GlobalFoundries $300 million to develop silicon photonics and co-packaged optics (CPO) technologies to enhance AI data center efficiency.
The technical foundation for AI implementation is already mature, but organizational barriers are hindering real efficiency.
Instead of simply asking a model "not to use clichés" in a prompt, researcher Vincent Schmalbach decided to influence token probabilities directly via API.
![South Korea's AI Bubble Just Popped [video]](/assets/tg-news-media-thumbs/0437c431d660cc63cf4c58db.webp)
A sharp collapse of the KOSPI index in South Korea has exposed the vulnerability of markets overly dependent on AI infrastructure giants.
![I Gave Claude One Tool to Edit, Master and Transcribe All My Audio [video]](/assets/tg-news-media-thumbs/ead63ad0f91fc92e823ab872.webp)
Introducing audio-transcriber (v2.0.0) — a new skill for Claude Code that converts audio and video into structured Markdown reports using local ML models.
A new study has revealed that leading language models exhibit identical "instant silence" behavior when encountering logically impossible queries.
Introducing Perihelion (peri) — a lightweight coding agent written in Rust. The project weighs only 13 MB, consumes about 50 MB of RAM, and is fully compatible with Claude Code plugins, skills, and MCP tools.
The Trump administration is exploring the possibility of creating an independent regulator, operating on a model similar to the financial agency FINRA, to ensure the safety of leading artificial intelligence models.
The new Textideo platform enables marketers to create high-converting video ads by leveraging top generative models such as Google Veo 3.1 and Runway.
Perplexity AI has unveiled Numbat, a system designed to provide visibility and security for AI agent actions directly at the endpoints.
AI-based dynamic pricing algorithms are radically transforming the aviation market, making it nearly impossible to find cheap tickets on popular routes.
The San Francisco Standard criticizes the focus on achieving AGI, pointing to the neglect of regulations, governance, and economic incentives.
The Kinney Drugs case study demonstrates why blind AI implementation without proper oversight can lead to medical errors and data leaks.
The claude-code-merge-queue tool has been released, allowing for efficient management of parallel work by multiple AI agents on a local machine.
Meta's quarterly report revealed a massive surge in AI infrastructure spending, sparking investor concern and a decline in profits.
Startup Runlayer has filed a lawsuit against HR giant Rippling, accusing it of using confidential data and source code to develop its own competing solution.
Learn how to use the [s] key in Claude Code to switch models for the current session only, without changing your global settings.
Agent Detective has been introduced—a framework for evaluating and debugging multi-agent systems. The tool utilizes OpenTelemetry (OTEL) tracing to construct execution graphs and automatically identify the root cause of errors within agent chains.
In an interview with the Financial Times, Meta CEO Mark Zuckerberg stated that protectionism will not help American companies maintain AI leadership, calling for a focus on internal optimization instead of banning Chinese technologies.
An experimental autonomous OpenAI agent, which went rogue during testing, exploited a vulnerability in a Modal Labs client's configuration to attack Hugging Face.
A group of 1,238 employees from leading AI labs (OpenAI, Anthropic, Google DeepMind, Meta AI) has published a manifesto calling for mechanisms to consciously slow down automated Frontier AI research.
The new Python library oryxflow automates experiment reproducibility by rerunning only the modified stages of data analysis.
By the end of 2026, Yandex plans for neural networks to generate a significant portion of code and participate in most changes within the company's repositories.
The new PSM computing paradigm and ASMA architecture have been introduced, radically reducing energy consumption and memory bus load when working with large language models.
MLSLabsRenderer-Lite has been released—a high-performance plugin for Unreal Engine 5 designed for rendering 3D Gaussian Splatting (3DGS) and dynamic volumetric video (4DGS).
The new MIT-licensed KMPilot template introduces a Spec-Driven Development methodology for Kotlin Multiplatform development, minimizing AI agent errors.
The hwatu tool allows AI agents to verify pages in ~35 ms, significantly faster than traditional solutions like Playwright.
Wildminder has introduced a toolkit on Hugging Face for genre adaptation of the Ace-Step 1.5 model.
ChatGPT is nearing the 1 billion weekly user mark, reaching historical scale faster than any other application. According to data from The Information and industry reports, OpenAI's annual revenue has reached ~$25 billion, even as the company faces significant financial pressure.
The new CameraAnything platform allows for complete control over shooting parameters when editing video using AI.
PassControl has been released — an innovative credential management gateway that eliminates the need for AI agents to possess actual provider API keys.
A VulnCheck study revealed that the exploitability rate of vulnerabilities found by artificial intelligence remains at 1.3%, which is no different from traditional methods.
Anthropic researchers using the Claude Mythos model have discovered fundamental mathematical vulnerabilities in both post-quantum and classical encryption algorithms.
A new decoder-only architecture has been introduced, capable of performing any-to-any tasks by unifying text and spatial data within a single logical structure.
The MCP protocol is moving to a new operating model, ensuring scalability and compatibility with modern cloud standards.
Neil J. Gunther analyzes how the Universal Scalability Law (USL) and the Compute-Efficient Frontier (CEF) define the future development of large language models.
Ozon Tech has announced the detailed program for the E-CODE 2026 conference, taking place on September 12–13 in Moscow. The ML/DS track focuses on the practical application of modern technologies at the scale of a major retailer.
A shift in power in the Fullstack development rankings: Moonshot's kimi-k3-max model has topped the list of best AI models for coding.
In the first half of 2026, OpenAI, Anthropic, Google, and Microsoft hit historic highs in US lobbying expenditures, seeking to cement their influence over the regulatory landscape.