@Sam

We’re pleased to announce the stable release of Qubes OS 4.3.1! This patch release aims to consolidate all the security updates and bug fixes that have occurred since the previous stable release. Our goal is to provide a secure and convenient way for users to install (or reinstall) the ...

(FWIW to you, dear reader, I did not use AI when writing this article) AI cannot desire. It can plan, and it can execute, but it cannot desire. If you desire X, you can use AI to come up with a plan to get X, and depending on the task, AI can execute parts or even all of the plan. But AI cannot want X. AI can plan an app and can build one, but it cannot want the app to be user-friendly/fast/follow a certain aesthetic/etc. You can want it to be all of those things, and if you do, you must tell it to plan for those things, because it cannot want to achieve them itself. What am I getting at here? AI is getting very good. I ask myself: What is becoming less valuable because AI can do it better/faster/cheaper than humans, and more importantly, what can't it do better/faster/cheaper than humans? I think the answer is desire. AI cannot desire. Some people have said that taste is what we bring to the table. I don't think that's wrong, I just think it's a downstream effect of desire. One has taste because they know what to look for, they have values, and they desire that the thing they're evaluating has lives up to those values. You get out what you put in. If you don't tell AI what to do, it will not think of it for you. It's not like asking a human to build or write something, because a human has its own goals, motivations, values, experiences, and ends. Get good at communicating your desires in writing. Cultivate your values. Learn to express them. It is one thing to have taste and values, it is another to be able to be able to express them in words. Criticism is valuable.

How and Why VCs Need to Update Their Playbook

The Envelope Collective ran from 2005-2007 and featured mail art sent in from all over the world. While we're no longer accepting submissions, the site serves as an archive of all of the wonderful art received during those years. Explore the gallery to see every single item we received through the span of this exciting project.

Google says YouTube’s terms of service grant a “broad license” for AI models to be trained on music uploaded to the platform in response to a lawsuit.

We propose inverse rubric optimization (IRO): tasks where an agent must learn the preferences of a black-box judge under a label budget. IRO tasks induce rich agent behavior and smooth scaling, making them a useful testbed for agent science.

Introduction

Curated resources to help you manage and govern your enterprise-scale agents with Agent 365 — the control plane for AI agents.

🩺 Introducing: VIBE ROUNDSVibe coding changed software. It's time medicine had its equivalent.Vibe Rounds is a concept for AI-augmented clinical care —...

The main website of the Data Systems Group (DSG) at MIT.

Automation should scale clarity, not replace judgment.

Estimating the value of progress with a rough calculation and a new web tool

A blog mostly about Clojure programming