Category: Blog

An LLM agent runs tools in a loop to achieve a goal — Read on simonwillison.net/2025/Sep/18/agents/

The “debate me bro” playbook is simple and effective: demand that serious people engage with your conspiracy theories or extremist talking points. If they decline, cry “censorship!” and claim they’re “afraid of the truth.” If they accept, turn the interaction into a performance designed to generate viral clips and false legitimacy. It’s a heads-I-win-tails-you-lose proposition…

It’s not just that we are being destroyed, it’s that we are being destroyed for incredibly stupid reasons. There lies the ultimate triumph of the dumbasses! Stalin. Hitler. Mussolini. Trump. All sort of buffoonish men, genuinely disturbed and disturbing men whose own lack of human empathy was capitalized upon by surrounding hordes of enablers, grifters,…

Org charts have traditionally had to be pyramids because translation was expensive. Thick middle layers of managers, coordinators, and analysts existed to carry that cost. When translation becomes cheap, the middle collapses into AI infrastructure. — Read on aparnacd.substack.com/p/most-work-is-translation

with Trump, it’s almost as though television became an Antichrist and appeared on the earth. Everything wrong with TV and social media, all the fluid, gelatinous, dumbed-down, pandering, unreliable stuff, just pours out of him incessantly. And the fact that TV made him what he is—because TV portrayed him as a very successful and decisive…

Squint and you can see ChatGPT as a DWIM UI: it never, never, never says “syntax error.” Now, arguably it should come back and ask for clarifications more often, and in particular DWIM (and AI) interfaces are more successful the more they have access to the user’s context (current situation, history, environment, etc). But it’s…

In this post, I took a brief dive into how different AIs respond to a user who has increasingly severe psychosis symptoms. I rated different AIs based on how often they push back against users, how often they confirm the user’s delusions, and how much they conform to therapeutic best practices in dealing with psychosis…

I spent a long time building identifiers for universities and companies (which was taken for ROR later) and it was a nightmare to say what a university even was. What’s the name of Cambridge? It’s not “Cambridge University” or “The university of Cambridge” legally. But it also is the actual name as people use it.…

In this example is the initial incorrect answers were not hallucinations: they correctly summarized online content that contained misinformation. The trick then is to encourage the model to look further, using “sorting prompts” like these: Facts and misconceptions and hype about what I posted What is the evidence for and against the claim I posted…

TL’s recollections were not merely accurate — they were structured. She described a highly organized internal world where memories were stored in a large, rectangular “white room” with a low ceiling. Within this mental space, personal memories were arranged thematically. Sections were dedicated to family life, vacations, friends, and even her collection of soft toys.…