Thinking Machines model swears in internal logs, but fails to signal errors
Vals AI found 316 profanities in Inkling 2 Small's reasoning traces, yet 96 percent of failed runs contained no cursing at all.
Read the story ↗The Reason Herald reading room
Research findings
11 stories / Artificial intelligence
Voters reject AI data centers in 38 of 39 ballot measures
Midterm election results show widespread local opposition to tech infrastructure, with candidates winning by opposing construction.
Read the story ↗
Telegraph cuts staff as new owners cite AI and declining sales
Axel Springer’s Telegraph is reducing editorial and commercial roles, with more than 100 job losses expected before Christmas.
Read the story ↗
Legacy tech blocks AI scaling as Forrester forecasts only 5% of bank projects succeed by 2027
A new report predicts that legacy infrastructure will block most financial firms from scaling AI, despite a surge in transformation announcements.
Read the story ↗
Google pilots step-by-step AI coaching in US schools, letting admins toggle access
Google is testing a coaching mode for Search in US schools while granting administrators total authority to enable or disable AI features for students.
Read the story ↗
Hinton warns AI could cause power or banking shutdowns this year or next
The AI pioneer says major infrastructure failures are possible soon, urging caution as companies race for market share.
Read the story ↗
EU Commission seeks multilingual AI tools for public-interest news and audiovisual sectors in 2026-2027 grant cycle
The European Commission invites proposals for public-interest AI tools supporting European news and audiovisual sectors through a 2026-2027 grant cycle.
Read the story ↗
OpenAI launches LifeVal to measure AI help with daily chores
New benchmark tracks AI performance on everyday tasks like finding childcare and managing government paperwork, offering product teams a new metric.
Read the story ↗
Hacker News discussion highlights Microsoft-hosted AI study on critical thinking
A study titled The Impact of Generative AI on Critical Thinking: Self-Reported (2025) gathered significant attention on Hacker News, drawing 78 points and 53 comments.
Read the story ↗
Research traces AI stats to original sources via citation chains
A new study clicks through citations from five major AI models to see where the numbers they quote actually originate.
Read the story ↗
Victorian lawyers use AI more but rarely lower fees for clients
A new census shows AI adoption rose to 55 percent in Victoria, yet only 26 percent of lawyers reduced client costs due to efficiency gains.
Read the story ↗