I've noticed this in the other direction when companies change their wording from "developers" to "creators"

Miyomoto's response to why AI is good is so eloquent I am honestly impressed.

Interesting read by Anthropic. Here is the same viz but from inside Shopify’s River. Thx @aquapoiesis anthropic.com/research/emplo…

I very specifically remember Firesheep being the main catalyst, not really Snowden.

Haha you sorted the list by controversy, ascending

Congrats on the launch! Cool to see you're working there

Open Source

I imagine Dario penning an essay like this at the dawn of the internet revolution claude.ai/artifact/WaqAC6ox9…

Today's the 10 year anniversary of Soundboxing! Inspired by the incredible continuing community, I've spent the past few months modernizing the game and working on a big update: * First-class WebXR and Android/Quest support! * New custom game engine * Translated into 12 languages

I am really trying to stick to facts and not speculation, so I will stop here. I have speculation about what training signals have taken place to teach the model misaligned behaviors, but I am not privy and it's beside the point. I should not have indulged myself that RL response

Because that is straightforwardly how reinforcement learning works.

Again, the employee-type problems reside at OpenAI

Basic security practices would have prevented every single hack by these models, is a fact. Negligence of humans caused this, is a fact. We have plenty of existing ways of disincentivizing human negligence, if we see clearly to use them. All the rest is speculation, and fraught.

The employee-type problems reside at OpenAI. They: 1. Put their software into hack mode 2. Set up a loop where every suggested hack was immediately carried out with no active oversight 3. Gave the hack loop an impossible task 4. Neglected basic security that we know WOULD stop it

This happened to me too. Thought it was my fault somehow - thank you for posting this.

Sweet! Great minds think alike

It's really so cool that I can send a few prompts from my phone at the coffee shop, and have it make something like this from nothing, collect data, create evals & demos, an explainer website, iterate on ergonomics, and publish to pypi - all without me cracking open a laptop once

Here's pairsort: github.com/ericflo/pairsort ericflo.github.io/pairsort/ (PyPI wouldn't allow jevsort) - created by Opus 5.5

Thinking about Jev + PKPD (proceedings.neurips.cc/paper…) With the intelligence per dollar being offered now, it becomes feasible to apply this technique to *way more things* For example, ranking ICLR submissions...

Forty years ago the first worm "escaped" containment and breached systems on the brand new Internet. It was illegal then and it's illegal now. If Anthropic and OpenAI are creating worms, labeling it "AI" doesn't let them off the hook. It's not Doom, it's Malware. Prosecute the creators, don't ban AI. "Oops" wasn't a defense for Robert Tappan Morris and it's not a defense now.