sum things considered v 1.1
Staying on top of the miscellany.
I’m growing fond of the idea of ‘lego blocks of thought’. This cluster of links are about the edges of the AI conversation. From data centers to underlying global trends from prompt hacking to context engineering, these links will expand your view of our new technology.
Making AI safe happens on a number of simultaneous levels. You can imagine each of these stories to be about an aspect of AI safety.
When a Billion Dollars Isn’t Enough: Why Mira Murati Is the Leader AI Desperately Needs.
‘The data is shocking: an LLM with the same prompt 1,000 times under identical settings produced 80 different outputs → structurally distinct completions with different reasoning paths.’
Murati and team develop tools that ensure answer consistency. Next horizon? Getting the right answers consistently. Follow Stephen Klein on Linkedin.How Data Centers Actually work.
How does a query on ChatGPT end up in a data center? Step by step.10 Forces Shaping Global Business. by the Boston Consulting Group.
Joey Price pointed this one out. In our rapidly changing world, geopolitics is having a larger impact on talent and HR than we’ve seen before.Pliny the Prompter.
A legendary repository of all of the kinds of LLM hacks from “The most prolific jailbreaker of ChatGPT and other leading LLMs.” Security is going to be the biggest issue in the development of LLM technology. Imagine a bank that keeps its doors open for robbers. Staying informed about the evolving security threats is a necessary thing in the next four or five years.Beyond Prompting: The Emerging Discipline of Context Engineering
LLMs are probabilistic. They follow patterns. If you want reliable, enterprise-grade output, the model must be placed in a structured environment. Context engineering creates that structure.We let OpenAI’s Agent Mode surf the web…here’s what happened.
A first look at the agentic browsing capabilities of OpenAI’s new browser. Are you ready for agents to continually visit your web property checking for updates?



