Good morning, Two Minute Warriors.

AI is changing how we work. Not all of it is good. We're becoming confidently wrong more often, leaning on AI's words instead of our own. But here's the thing: if you're going to trust AI anyway, use it to sharpen your human connections, not replace them.

⏱ The 10-Second Version

  • A study from three European universities found that access to AI advice dropped accuracy from 27% to 9% while confidence jumped from 30% to 76%. People became worse and had no idea.
  • The effect wasn’t just over-trust: AI suppressed the habit of saying “I don’t know,” collapsing that response from 44% to 3% even when participants would have gotten the answer right on their own.
  • A separate study found ChatGPT’s signature words (“delve,” “meticulous,” “showcase”) are now measurably spreading into unscripted human speech, confirmed across 737,000 hours of podcast audio.

The Big Thing

AI didn’t just make people wrong. It made them confidently wrong

What happened

Researchers from École Normale Supérieure, University of Milano-Bicocca, and Sapienza University of Rome ran participants through questions where the AI model (Step 3.5 Flash) was typically wrong. With AI access, accuracy fell from 27% to 9% and willingness to say “I don’t know” collapsed from 44% to 3%. Confidence rose from 30% to 76%. Adding monetary incentives barely moved the needle.

Why you care

The risk isn’t that your team trusts AI. It’s that AI quietly switches off the habit of flagging uncertainty. Some participants who would have answered correctly on their own asked the AI and got it wrong. The mechanism is cognitive, not informational.

What to say in your next meeting

“The study worth flagging isn’t that AI gives wrong answers. It’s that it kills the instinct to say ‘I don’t know,’ even in people who would have gotten it right without it.”

Speed Round

A study of 25,264 agentic PRs across 2,361 GitHub repos found that median adoption is still 1–2 agentic PRs per repo per quarter. Intensive use is concentrated in a tiny share of projects.
Read the research →

A preregistered study of 737,083 hours of podcast audio found that ChatGPT’s signature words (“delve,” “meticulous,” “showcase”) causally increased in spontaneous human speech after the chatbot’s release.
Read the research →

New York City Mayor Mamdani announced landlords must disclose when AI-generated images are used in rental listings, the city’s first rule of its kind.
Read the full story at PetaPixel →

New research across 12 frontier models and seven benchmarks found that fixed-budget AI evaluations increasingly understate real capability. Scores are protocol-dependent, not just model-dependent.
Read the research →

The Two-Minute Win

A research brief in 2 minutes, not 35

Before

30–35 minutes reading a study, pulling the key numbers, writing a summary, and figuring out what it means for your team.

After

Under 2 minutes.
  1. Drop the URL or PDF of the study (today’s example: the cognitive surrender paper) into Two Minutes AI.
  2. Two Minutes AI reads it and produces a structured brief: what happened, key numbers, and what it means for your work, ready to share.
  3. Skim the output for one line to adjust. Forward or drop it into your doc. Done in under 2 minutes.

Inside Two Minutes AI: Works on any plan. Paste a link or upload the PDF directly.

Prompt of the Day

The Uncertainty Extractor

“Read the following AI-generated output and do two things: (1) list every claim that depends on the AI being correct about a specific fact, and (2) flag any place where a human expert would likely say ‘I’d need to check that.’ Don’t fix anything — just surface the assumptions. Output: [PASTE AI OUTPUT HERE]”

Why it works: AI outputs are written to sound complete. This prompt forces the gaps to the surface before they get copy-pasted into a document or decision.

Time saved: 20 minutes of manual skepticism becomes a 90-second checklist.

Two Minutes AI builds source-grounded briefs that flag assumptions by default, no prompt required.

Forward this to the one person on your team who reviews AI outputs by nodding along.

Don

Keep reading

View more
caret-right
© 2026 Two Minutes.
beehiivPowered by beehiiv