Browse Comments — Raw (as collected)

The weakest layer in many companies is usually not the LLM. It is the connection layer: permissions, tool access, data boundaries, and reliable integrations.

100%. Even when I prove it wrong or that it contradicts itself, at best it just acknowledges my point and nothing happens. Does it actually learn?

This is hilarious

Scott Liddell

One aspect that stood out to me was the distinction between governance as a checklist and governance as a process of discernment. Too often, conversations about AI focus on rules, compliance, and control. But the deeper challenge is cultivating the wisdom, responsibility, and human judgment needed to apply those frameworks well. I also appreciated the emphasis on education. If AI is reshaping society, then schools aren’t simply preparing students for jobs—they’re helping prepare citizens to navigate a world increasingly influenced by intelligent systems. Thoughtful reflection like this is exactly what the AI conversation needs more of. Matt Davis Founder/CEO CivicTruth Media Group Founder & CEO The Patriot Party Movement Founding Partner & Civic Educator American Institute for Civic Leadership

As Demis Hassabis points out, breakthroughs in world understanding and agentic systems could redefine how humans interact with technology and tackle complex challenges in the years ahead. 🌍🚀

It's 'maths'

The detail that should worry people most is that the severity score held at 8.0 across every language. The model understood the danger. It just routed that danger into a different action based on a location nobody asked it to assume. That kind of failure passes a clean English eval and only surfaces once real users hit it. Anchoring location fixes this case, but the bigger lesson is that evals have to hold everything constant except the variable you are testing, or the divergence stays invisible until production.

Tianyu Xu / Mr. AI-Video, very interesting! How can we generate pictures from rooftops and buildings in high resolution? Often are digital drawings not available and Google Earth isn’t helpful…

SUD/MH Voice AI Admissions Assistant DIAL3D

In fact this is a sad situation and it is lose-lose.

The issue isn’t what AI can do. The issue is that governance keeps reacting instead of governing. If the structures responsible for oversight don’t set the terms, the terms get set by default.

🤣

Binyamin Klempner Thank you for your clarification. I apologize for my misunderstanding. "Shared discernment" when it comes to drafting a papal encyclical necessarily refers to the process of gathering information and drafting. In this case, the Pople likely consulted with a wide variety of secular and religious experts to gain an understanding of the subject matter and the issues raised by AI's ascendance. It the end, though, the encyclical speaks with his voice because it's his name on the encyclical. His voice then contributes the Church's perspective to the larger global effort of "shared discernment" as it struggles to answer the questions raised by AI in a sensible way.

Pascal BORNET I love your posts. They're always funny and controversial enough to make you stop and think. Case in point the video above is hysterical. Clearly, I appreciate dark humour. But then I got to the line about AI being designed to make humans more capable, more creative and more central to the future we’re building - and honestly, it made me sad. I mean that woudl be Awesome! I want to believe that’s the future we’re building. I really do. And I've always been optimistic about what AI can do for people, especially those who've historically had less access and fewer opportunities. BUT it hit different today.... the scales don’t feel balanced. And that's actually not funny it's quite frightening. When so much of the power sits with so few people, it’s hard not to question what future is actually being built, who its being built for, and how much say the rest of us really have.

Osama Sohail another way: ask AI to criticize itself. E.g. “Read this like you are evidence based medicine expert…” when you discuss a medical topic.

It's like asking a 10 year old how confident he is about string theory!! In his mind, the kid will always be confident and keep giving you more and more hallucinating/made up answers.

Pascal BORNET I believe this is a RAG issue that needs to be debugged.

Exactly! Well put! To my understanding the process of composing the encyclical is a collaborative process employing a variety of experts in the given field in which both the subject matter at hand is examined as well as how that subject matter fits, or doesn't fit, into a theological framework. The main examination being how the subject at hand adds to, or detracts from, theology, which, essentially is the study of man's relationship with the Divine. At least that's my understanding.

Is this not an episode of Philip K. Dick's electric dreams?