Browse Comments — Raw (as collected)
Close reading of the corpus at each pipeline stage: raw → clean → relevant → coded.
4.3K
comments matched
· page 196 of 215
Curiosity is probably the single best antidote to AI overconfidence, the instinct to ask one more question before accepting the output is underrated as a workplace habit
Teams that build a habit of asking "what would make this wrong" before acting on AI output are going to make sharply better decisions over time
The confident-wrong pattern also trains people to distrust their own correct instincts when they conflict with what the AI said, that's a subtle and serious problem
There's something almost philosophical about a tool that remains equally confident whether it's describing photosynthesis or inventing a citation that doesn't exist
The breathing exercise redirect is going to be a case study someday in the gap between user experience design and information integrity design
The best leaders right now are the ones who've made "cite your source and show your reasoning" a team norm that applies equally to humans and AI outputs
What's interesting is that humans who are confidently wrong usually have a tell — a slight pause, a hedge — AI has no such tell, which changes the verification burden entirely
Something about the way AI never says "actually, on reflection" makes the humans around it more responsible for building that reflection step into the process
This is exactly where AI coding becomes a real skill. Claude Code can move fast, but the quality still depends on how clearly you break the system down, review each layer, and test the weak spots.
Pascal BORNET The bear is dangerous. A confident hallucination telling you everything is under control might be even more dangerous. 😄
}
Hello Demis,
I hope you are doing well.
I have been following the progress of DeepMind for many years, and I find your work at the intersection of AI, neuroscience, and complex problem-solving truly remarkable.
I am Dr. Zaid El-Junidi, Founder of El-Junidi Engineering Group in the United Kingdom, working in architecture, engineering consultancy, and interdisciplinary project development.
I am particularly interested in how AI systems are shaping decision-making, design processes, and the future of built environments.
It would be an honor to connect and stay in touch.
Kind regards,
Dr. Zaid El-Junidi
El-Junidi Engineering Group
United Kingdo
You should talk with Sharon Zhang and Suman Kanuganti their personal.ai product would be a perfect match for your framework.
I plan to wash my car at a car wash located 20 metres from my home. Should I walk or drive?
Francis E. I think we may be a little way off from this. Unless theses data centres will be completely unmanned I’m unsure Mr Bezos has taken into account the human factor, an engineer has a family emergency and needs to get home. Or maybe an agonising wisdom tooth which requires treatment?
The training of engineers to become astronauts, returning from rotation and performing a MIR space station style docking procedure when it’s challenging enough to get an engineer who lives in Italy to Northen Sweden.
But your comment has fired my imagination, thank you.
Rayna Sweet why do I see misspelled words all the time? Put some spell check on your comments or check your AI.
Justin, I will suggest the absence of self awareness.
There's a version of leadership development that should just be: practice disagreeing with things that sound authoritative, AI outputs make great training material
This is exactly why AI won’t replace human judgment anytime soon.
It can process patterns insanely fast, but it still lacks something very human like situational instinct.
A bear charging at you is not the moment for “balanced emotional guidance.”
It’s the moment for survival-driven clarity.
The danger isn’t AI being wrong but humans are wrong too.
The danger is how confidently and fluently it can sound right while missing context completely.
There is a generation of junior employees forming their research instincts right now in environments where confident-sounding output is the norm, that's a slow-burn problem worth naming
David Nour a question if I could about the internships. Are these paid roles?