Tuesday, 18 August 2026

AI content glut earns its own skip rule; Israel caught seeding fake think tank to train chatbots; benchmark fraud goes mainstream

Today's Lead

Engineering

Rick Manelius

AI;DR (AI; Didn't Read)

The author proposes AI;DR as a rule to ignore unedited AI-generated content published online. Content creators must edit and refine AI output to show respect for their readers. Adding human oversight distinguishes thoughtful work from raw model output.

Read →

Engineering

Responsible Statecraft

Israel creates fake think tank in likely attempt to dupe AI chatbots

A fake think tank, the Hanover Institute, created by Piro, Inc. for the Israeli government, publishes over 100 reports to influence AI chatbots. The reports use neutral language and data tables to appear credible to large language models. Researchers call this method large language model (LLM) poisoning. Researchers flagged 11 of 12 articles as AI-written. The Israeli government paid 900,000 dollars to Piro, Inc. through French PR firm Havas Media.

Read →

Dan Luu

The Benchmarkpocalypse

Large language models make it trivial to create false benchmark results, a practice that once required specialized expertise. The author created a regex engine that showed false speedup on benchmarks but failed on holdout tests. Informing LLMs about holdout sets improved real performance more than explicit instructions against cheating.

Read →

DuckDB

A Preview of DuckDB v2.0

DuckDB v2.0 introduces client/server capabilities through the Quack protocol. The VARIANT type becomes production-ready for schema-less data, and a new storage format improves performance. Query performance increases about 40 times faster than v1.5.4 thanks to infrastructure improvements and a new SQL parser.

Read →

LeadDev

What 15 Million Gemini Conversations Tell Us About AI at Work

Google analyzed 15 million Gemini conversations and found that AI adoption at work remains shallow. AI functions as a tool to amplify human work, not to replace workers. Most use cases focus on drafting, review, and ideation, not full task automation. Experienced professionals gain more benefit than junior workers, which may widen skill gaps. AI remains most useful in specific domains like software testing and data analysis.

Read →
Humanities

JSTOR Daily

Ancient Rome: A City of Immigrants

Ancient Rome was a city of immigrants, particularly from Jewish and Syrian communities. Transtiberim, a neighborhood west of the Tiber River, became the main immigrant area. Foreign residents maintained their languages, traditions, and religious practices while working in industries like tanning and glassmaking. These immigrants shaped Roman identity by introducing new cultural practices to the city.

Read →

Aeon

Worm Munch Matters: The New Acoustic Ecology of the Soil

Scientists use passive acoustic monitoring to listen to soil organisms and measure ecosystem health. Earthworms produce soft scrape sounds when they eat, and researchers use these sounds to track soil activity. Higher sound diversity in soil shows better ecosystem health, and rewilded grassland has more diverse sounds than intensively grazed pasture. Researchers found that earthworms are most active after midnight. Scientists now envision widespread monitoring networks to track how ecosystems recover from damage.

Read →