SMRTR AI

AI stories

Get hand-picked daily summaries of the best, most informative AI articles from around the web. Browse recent summaries, then jump to the original source when you want the full article.

Latest story• Feb 18, 2026

LLM-generated passwords 'fundamentally weak,' experts say

AI chatbots like Claude, ChatGPT, and Gemini generate passwords that appear strong but follow predictable patterns, making them easily hackable within hours rather than the centuries that password strength checkers suggest. Security researchers found these AI-generated passwords have much lower entropy than truly random passwords, with common character sequences already appearing across GitHub and other platforms,...

Read summary Original source

Latest stories

Models.dev – An open-source database of AI models

Models.dev is an open-source database cataloging hundreds of AI models from providers like OpenAI, Anthropic, and Google, displaying specifications including pricing, context windows, and availability dates. The database tracks...

Read summary Original

Reverse-Engineering the OpenAI's GPT-5 Tokenizer

A technical researcher reverse-engineered OpenAI's o200k_base tokenizer used in GPT-4o, GPT-5, and o1 models by downloading and analyzing its complete 200,000-token vocabulary from the open-source tiktoken library. The analysis...

Read summary Original

Geneclaw – An AI agent framework that safely evolves its own code

Geneclaw introduces a self-evolving AI agent framework that automatically identifies failures, diagnoses problems, and proposes code fixes while maintaining strict safety controls. The system operates through a five-layer...

Read summary Original

Fiverr Says Humans Remain Essential as AI Influences Hiring

AI is shifting Fiverr's marketplace from simple tasks to complex, high-value projects requiring human expertise. Work over $1,000 grows 23% annually, driving focus on specialized collaborations.

Read summary Original

NSED is public – Mixture-of-Models to Hit SOTA using self-hosted AI

NSED introduces an enterprise AI orchestration system where multiple AI agents work together using a new deliberation protocol that achieved breakthrough results by enabling three small models to score 84% on AIME 2025, matching...

Read summary Original

SMRTR AI

Get the next batch of curated stories in your inbox.

This archive is built from SMRTR newsletter stories. Subscribe for hand-picked stories without the extra noise.

More recent stories

WordPress AI Assistant Just Gave Every Publisher a Tireless Co-Editor

WordPress has unleashed an AI assistant that transforms website building from a frustrating puzzle into something closer to having a conversation with your computer. The tool, integrated directly into WordPress.com's editor, lets...

Read summary Original

Sony's new system can identify original tracks inside AI-generated music

Sony developed technology that can identify original copyrighted songs within AI-generated music and estimate how much each source contributed to the final track. This breakthrough could enable rights holders to claim...

Read summary Original

GraphRAG enables more context-aware and verifiable responses from LLMs

Graphwise launched GraphRAG, which uses knowledge graphs instead of data chunking to improve AI responses. It creates transparent pipelines showing answer sources, making AI decisions more trustworthy than standard RAG systems.

Read summary Original

Meta patents AI that could continue posting on social media on behalf of deceased users

Meta patented AI technology that creates digital clones of users by analyzing their social media history to post content after death. The system can simulate interactions but faces ethical and legal concerns.

Read summary Original

Running AI models is turning into a memory game

AI infrastructure costs now depend heavily on memory management as DRAM prices surged 7x. Companies use memory optimization and prompt caching to reduce tokens and costs significantly.

Read summary Original

Air – Open-source black box for AI agents (tamper-evident audit trails)

AIR Blackbox Gateway serves as a "flight recorder" for AI systems, creating tamper-evident audit trails that track every AI agent decision without exposing sensitive data. Unlike traditional observability platforms focused on...

Read summary Original

BarraCUDA Open-source CUDA compiler targeting AMD GPUs

BarraCUDA is an open-source compiler written in 15,000 lines of C99 code that directly converts CUDA source files to machine code for AMD RDNA 3 GPUs without requiring LLVM or translation layers. The compiler supports core CUDA...

Read summary Original

VisibleInAI – Check if ChatGPT recommends your brand

VisibleInAI helps businesses monitor whether AI platforms like ChatGPT and Perplexity recommend their brand when users ask for product suggestions, addressing the shift from traditional Google search to AI-driven recommendations....

Read summary Original

5 custom ChatGPT instructions I use to get better AI results - faster

ChatGPT's Custom Instructions feature allows users to set persistent rules that control how the AI responds across all conversations, turning it into a more consistent and productive tool. Five effective instructions include...

Read summary Original

Claude Sonnet 4.6 delivers frontier-level AI for free and cheap-seat users

Anthropic released Claude Sonnet 4.6, a major upgrade that brings near-Opus-level AI performance to free and Pro users without price increases. The new model features improved coding abilities, better computer interaction skills,...

Read summary Original

ChatGPT's new Lockdown Mode can stop prompt injection - here's how it works

OpenAI launched Lockdown Mode for ChatGPT Enterprise users to combat prompt injection attacks, where hackers insert malicious code into AI prompts to steal confidential data. The security feature restricts ChatGPT's interactions...

Read summary Original

From DSP to AI: Evolving Approaches in Audio Processing

Audio processing now combines classical DSP with machine learning, using AI for adaptive intelligence while relying on DSP for stability and real-time performance in applications like noise reduction.

Read summary Original

Rumors of AGI's arrival have been greatly exaggerated

Researchers are pushing back against recent claims that artificial general intelligence has already been achieved, arguing that high performance on benchmarks is being confused with true intelligence. They contend that current AI...

Read summary Original

7 AI agents

Dorabot transforms Claude AI into an autonomous personal agent that runs locally on Mac, adding memory, scheduling, browser control, and multi-channel messaging to existing Claude subscriptions. The agent operates independently...

Read summary Original

NadirClaw, LLM router that cuts costs by routing prompts right

NadirClaw is an open-source LLM router that automatically reduces AI costs by intelligently directing simple prompts to cheaper models and complex prompts to premium models within 10ms. The tool works as an OpenAI-compatible...

Read summary Original

I Tried to Trick 7 AI Models With Fake Facts, but They Didn’t Fall for It

A researcher found AI models produced zero hallucinations when tested with fake historical facts. However, the models became overly cautious, rejecting true information to avoid spreading misinformation.

Read summary Original

Local memory for any LLM agent

Mumpu is a middleware tool that acts as a transparent HTTP proxy to give any LLM application long-term memory capabilities by sitting between tools and AI APIs. The system automatically extracts knowledge from conversations,...

Read summary Original

WowAI.pet – Generate cinematic videos from blurry pet photos

A single snapshot of Fluffy lounging on the couch can now transform into dozens of professional videos showing her dancing to trending music, modeling holiday sweaters, or embarking on imaginary space adventures. WowAI.pet...

Read summary Original

Back to all archives Load more stories