Measured public attention
The questions shaping AI now
Nine evidence desks follow the questions drawing the strongest measured attention. Every item carries a date and a route to its source.
Measured 11 October 2026
Latest measured news
The newest ten
500B Tokens Later: Letting AI Agents Decompile a First-Person Shooter
momo5502.com
AI is creating a 'human premium' for art created by people
www.bbc.com
Anthropic Bans Cruelty to Claude, Still Won't Say What It Protects
rews.cc
Build your own decision model
nishtahir.com
Talorys – A self-hosted personal AI agent on Cloudflare's free tier
github.com
Nvidia in talks to acquire US 'open' model startup Reflection AI
www.ft.com
Vibe coded browser ports of Halo, The Simpsons: Hit And Run, GTA work well
kotaku.com
My personal AI agent posted my bank details on company Slack
www.businessinsider.com
Typesafe AI raises $870M at $7.5B
typesafe.ai
Show HN: Let your AI agents paint big arrows, boxes and text on your screen
github.com
Which model is actually ahead?
Full report →Model releases are the single largest attention magnet of 2026: 60 release threads above 300 points, about 60,000 points in total.
Can you trust what an AI agent does?
Full report →Agents acting on their own — publishing, attacking, spending — produced the most-read incident reports of the year.
Do you need the big labs at all?
Full report →Local and open-weight models moved from hobby to default option in the discussion.
Who owns AI — companies, markets or the state?
Full report →Acquisitions, lawsuits and government access decisions drew as much attention as the technology.
Is this written by a person?
Full report →The most upvoted AI thread of 2026 is a rule against machine-written comments (4,229 points).
Will AI take my job?
Full report →The widest public concern worldwide, by survey.
Can AI discover something new?
Full report →Mathematics became the public test case in 2026.
What kind of thing is an AI?
Full report →A maker now describes its own model as a new kind of entity, and has written rules about how users treat it.
What does AI do with my data?
Full report →Silent installs and always-on devices trigger the sharpest reactions.
How attention was measured
Attention is measured, not guessed: every Hacker News story from 1 Jan 2026 with more than 300 points and an AI subject (631 stories), read through the public search interface on 11 Oct 2026, grouped by the question behind the story. Survey figures come from published polls with four-digit samples and are linked.
Independent AI research
What is actually proven about AI.
Models, agents, chips, regulation, markets: 3,258 claims from 2,484 named sources. Each one shows where it comes from, when it was recorded — and what that source does not prove. For anyone who has to decide about AI while the headlines still disagree.
Free
No sign-up
How a record works
Every claim shows its limit.
Most AI coverage tells you what happened. It rarely tells you how sure anyone can be. Each record here answers four questions, always in the same order.
- 01
What is claimed
One bounded statement. Attention, adoption, revenue and impact are never blurred into one word.
- 02
Who says so
The original source with publisher, date and language — linked, so you can check it yourself.
- 03
What it does not prove
The limit of the evidence, written as plainly as the claim itself, not hidden in small print.
- 04
What would change it
The observation that would move the verdict. You know in advance what to watch for.
Signposts
The questions we watch.
Each signpost is a question with a threshold set in advance, and a dated verdict: observed, emerging, contested or not observed. No invented probabilities. The most recent entries:
Eight subject areas
Start with your question.
Every area draws on the same set of records, so two pages can never quietly disagree about one fact.
Models & capabilities
What a system can actually do, as opposed to what its launch material says.
Agents & automation
Whether a system can run unattended for hours, not whether it answers well once.
Compute, chips & cloud
Where the limit is physical: silicon, power, cooling, data-centre capacity.
Data & research
How a number was produced, before anyone repeats it.
Safety & regulation
When an obligation becomes a product decision, a deadline or a liability.
Markets, capital & work
Where money and jobs matter more than capability.
Industries & use cases
Evidence from one industry instead of a general statement about all of them.
Geopolitics & society
Where the decisive actor is a state, a bloc or the public, not a company.
Sources
Where the evidence comes from.
Official and primary sources first. The map shows which countries have a checked official source — and leaves visible where we have none yet.
- Named sources
- 2,484
- Languages
- 13
- Regions
- 10
- Signposts
- 42
Official sources say more about what governments intend than about what they achieve. Courts, procurement and regional rules are underrepresented.
See every source →
Who publishes this
A company that builds AI systems — said up front.
Metatron Intelligence is published by Metatron Marketing Ltd., which also sells AI implementation to businesses. You should know that before you trust a single line here.
Subjects are chosen on their merits, and a record says what the evidence says even when that does not help anyone sell anything. Where a piece touches our own commercial work, it says so.
Editorial policy →