AI Peer Review Beats Humans + Unprecedented Bot Traffic Surge
A new Harvard study found AI-written academic reviews now outperform the best human peer reviewers on every tested measure. Separately, a security researcher logged 14,000 honeypot hits in 6 hours — nearly all non-human — signaling an alarming surge in automated threat activity.
What we know
An account on X (Sabine Hossenfelder) RTs a viral thread: Harvard study of 134 papers finds AI reviews beat the best human reviewers on every metric, reversing earlier findings that AI should only assist. On the threat side, another account on X RTs a security researcher with a decade of experience who watched their honeypot server get hit 14,000 times in 6 hours, 'almost none of it human' — calling it unprecedented. Another account on X added context: 'AI is creating the threats; cybersecurity is cashing in.' @toly (Solana co-founder) pushed back on security alarmism, arguing AI acceleration is actually *closing* vulnerabilities rather than opening new ones. Meanwhile another account on X flagged a Chinese lab paper enabling trillion-parameter LLM inference across separate datacenters.
Sources
“No loss of life or property damage from AI acceleration — existing vulnerabilities being closed”
View post →By the Global Eye 24 desk, with AI assistance. Updated as the story develops. Spot an error? Write to corrections@globaleye24.com.