AI Unfiltered

10 Min Read

onApril 1, 2026

DeepSeek Exposed 1 Million+ Chat Logs Through Passwordless ClickHouse Databases—Wiz Researcher Ran SQL Queries via Web Browser on January 29, 2025

A security researcher opened a web browser, typed a URL, and gained full SQL access to production databases containing over 1 million chat…

Discover More

8 Min Read

Artur Markus

onMarch 31, 2026

Gemini 3.1 Pro Preview Hits 94.1% on LM Council Benchmark—Google Takes Three of Top Four Leaderboard Spots as Claude 4.6 Opus and GPT-5.4 Trail by Double Digits

Google just claimed three of the top four spots on major AI benchmarks while charging 60% less than the competition. The Claude-GPT duopoly…

Discover More

9 Min Read

Artur Markus

onMarch 30, 2026

Anthropic’s Model Context Protocol Hits 97 Million Installs on March 25—MCP Transitions from Experimental to Foundation Layer for Agentic AI

Six months ago, MCP was a footnote in Anthropic’s documentation. Today it’s the plumbing underneath 97 million AI agent…

Discover More

11 Min Read

Artur Markus

onMarch 29, 2026

12 AI Models Launched in One Week During March 10-16, 2026—OpenAI, Google, xAI, Anthropic, Mistral, and Cursor Compress Developer Selection Cycles to Monthly as Frontier Model Releases Pile Up

Twelve AI models in seven days. Developer teams that used to evaluate quarterly releases now face monthly decision cycles—and most aren’t ready.

Discover More

9 Min Read

Artur Markus

onMarch 28, 2026

ByteDance’s DeerFlow 2.0 Hits #1 on GitHub Trending with 35,300 Stars in 24 Hours—SuperAgent Framework Executes Code in Docker Sandboxes, Not Chat Windows

ByteDance just shipped an AI agent that doesn’t suggest code—it writes Python, spins up bash terminals, and deploys web apps inside its…

Discover More

9 Min Read

Artur Markus

onMarch 27, 2026

Microsoft Unifies Copilot Teams on March 17—15 Million Users Represent Just 3% of Enterprise Base as Company Admits Lag Behind OpenAI and Google

Microsoft just published its own report card, and the grade is a C-minus. After two years and billions in infrastructure investment, Copilot…

Discover More

11 Min Read

Artur Markus

onMarch 26, 2026

Moonshot AI’s Attention Residuals (AttnRes) Achieves Same Training Loss with 1.25× Less Compute—New Transformer Architecture Replaces Fixed Residuals with Softmax Attention

Residual connections have been sacred architecture for nine years—and Moonshot AI just proved they were holding us back. Their new AttnRes…

Discover More

11 Min Read

Artur Markus

onMarch 24, 2026

Meta Locks 6.6 GW of Nuclear Power Through Vistra, TerraPower, and Oklo Deals—Largest Corporate Nuclear Commitment for AI Infrastructure

Meta just committed to more nuclear capacity than Switzerland’s entire grid produces. The AI energy arms race has moved past…

Discover More

9 Min Read

Artur Markus

onMarch 23, 2026

PromptEng 2025 Workshop Accepts 7 Research Papers on January 24—First Academic Forum Dedicated to Prompt Engineering Techniques

While engineers debated whether prompt engineering was “real engineering,” seven research teams quietly got their papers…

Discover More

10 Min Read

Artur Markus

onMarch 17, 2026

OpenAI Launches GPT-5.4 ‘Thinking’ with 1 Million-Token Context Window on March 5, 2026

OpenAI just shipped a model that can hold 2,000 pages of text in working memory. Most enterprise RAG pipelines are still chunking documents at…

Discover More

9 Min Read

Artur Markus

onMarch 16, 2026

University of Montreal Tests AI Against 100,000 Humans on Creativity—GPT-4 Beats 72% But Top 10% Still Win

GPT-4 now outperforms the average human on standardized creativity tests. But before you fire your creative team, consider this: when…

Discover More

7 Min Read

Artur Markus

onMarch 14, 2026

Allen Institute Releases SERA-32B: Open-Source Coding Agent Hits 54.2% SWE-Bench Score for $1,300

A 32-billion parameter model now outperforms its 110-billion parameter teacher after fine-tuning on just 8,000 samples. The cost to train it…

Discover More

9 Min Read

Artur Markus

onMarch 13, 2026

White House Drafts ‘Any Lawful Use’ Mandate for AI Firms—Anthropic Blacklisted as Pentagon Supply-Chain Risk After Refusing to Waive Ethical Red Lines

The U.S. government just made AI ethics a disqualifying condition for federal contracts. Anthropic chose principles over Pentagon dollars—and…

Discover More

DeepSeek Exposed 1 Million+ Chat Logs Through Passwordless ClickHouse Databases—Wiz Researcher Ran SQL Queries via Web Browser on January 29, 2025

Gemini 3.1 Pro Preview Hits 94.1% on LM Council Benchmark—Google Takes Three of Top Four Leaderboard Spots as Claude 4.6 Opus and GPT-5.4 Trail by Double Digits

Anthropic’s Model Context Protocol Hits 97 Million Installs on March 25—MCP Transitions from Experimental to Foundation Layer for Agentic AI

Explore Trending Topics

Subscribe to my Blog