#LLM

Scientists Show How RoPE Positional Encoding Shapes the "Fingerprint" of Attention Heads in LLMs

Scientists Propose TSF Method: LLM Improves Industrial Process Forecasting

ROAM Framework Adapts Industrial AI Models to New Scenarios Without Retraining

Claude Paraphrased a Document After Reading Only the Beginning: Hidden MCP Server Truncation

A developer caught Claude confidently paraphrasing a document it hadn't finished reading—blame a quietly truncated MCP server response.

Developer Built a Local Spam Filter with ruBert-base-antispam Model Without VPN

An author on Habr assembled a multi-layer spam filter: after Yandex, Mail, and SpamAssassin, emails are checked by a local BERT model with 177 million parameters.

US Commerce Department lifted export restrictions on Anthropic's Claude Fable 5 model

The US Department of Commerce lifted export restrictions on Anthropic's most powerful models, ending a three-week block on Claude Fable 5 access for the company's…

MTS AI opened the AVI code — an external filter that keeps LLM within the law without retraining

MTS AI posted on GitHub a working AVI service — a model-agnostic filter that checks LLM requests and responses for legal violations, toxicity, and ethics.

01Kimi K2.7 Code versus Claude Fable 5: landing pages 94% cheaperJul 19, 2026·2 min 02LLM agents and SOC: developers built a cyber threat database using RAGJul 19, 2026·2 min 03Did GitHub become more uniform after ChatGPT? Code study for 2019–2025Jul 19, 2026·2 min 04Claude Sonnet 5 vs Opus 4.8: agentic coding comparison and API pricing from AnthropicJul 19, 2026·1 min 05YuMoney Built RAG Platform and AI Assistant Eugene for Corporate Wiki SearchJul 19, 2026·2 min

Top themes

LLM·GitLab Blog

Same LLM Benchmark Yielded 28% and 76% — Difference Is Only in How Questions Are Asked

Symlink Vulnerability Affects Six AI Code Agents, Including Amazon Q and Cursor

Wiz found a Unix symlink trick that bypasses the protection of six AI code agents, including Amazon Q and Cursor, and gives attackers access to a developer's…

Anthropic Launches Claude Science — AI Work Environment for Scientists in Beta

Anthropic introduced Claude Science — a beta application that brings together research tools and assigns AI agents significant portions of scientific work.

Kontur engineer compared D&D wizards and warlocks to the LLM era on Habr

Kontur's senior QA engineer Alexander Ovchinnikov published an essay on Habr where the shift from wizards to warlocks in D&D is a metaphor for how LLMs are changing professionals' work.

LLM·Anthropic Blog

Anthropic released Claude Sonnet 5 — the most agentic model in the Sonnet line

Developer replaced repeated context explanation for neural network with LLM Wiki workflow

X Launches Hosted MCP Server for Direct Connection to Claude, Cursor and Grok Build

AI shopping outpaced search on Prime Day: Amazon received traffic from chatbots

An AI agent is an LLM in a loop in 20 lines of code: where complexity begins

Jul 19, 2026·1 min

Meta restricted engineers' use of Claude Code and Codex due to distillation risk

PWN AI expert Artyom Semenov: how AI agents delete files and compromise themselves

LLM·AWS Machine Learning Blog

Amazon Seeks Cheaper Claude Alternatives as Anthropic Moves to Token-Based Billing

Jul 19, 2026·1 min

AWS Demonstrated Five Resilience Patterns for Generative AI on Amazon Bedrock

AWS Machine Learning Blog Described Five Practical Resilience Patterns for GenAI Applications on Amazon Bedrock — from Native Service Functions to LLM Gateway

Google Released Android 17 with Gemini Intelligence and Gaming Mode for Foldable Smartphones

Google Began Rolling Out Android 17 on Pixel: Gemini Intelligence AI Features, Separate Gaming Mode for Foldable Devices, and Stricter App Privacy Settings