How's things here? What's new in the last few months?

Petition to call it “a suit that you look very dashing in if I do say so myself”

Calling it a “harness” is probably not the best move for alignment purposes.

Learned about ARC-AGI-3 the other day, but only just tried it out. It's somewhat fun: arcprize.org/tasks/ls20. From what I recall, LLMs score very poorly on this vs humans, because the goals are not clearly laid out. So success means intuiting the goal. Honestly I prefer my LLMs to not have an innate motiva... Show more

OK, hear me out... Memento as an allegory for the an LLM's experience.

> AI won't take your job, humans using AI will take your job. I generally reject the meta that you can win by being better at using and directing AI than everyone else. If AI is actually good, and actually brings abundance then this win/lose scarcity mindset is obsolete. Anyone who is actually confident that AI will... Show more

exhibit B

The underclass will be permanent. The upperclass not so much.

It really speaks to the quality of Apple hardware, that after all these years, my iPhone 12 seems to charge faster than it ever has before.

One of the most powerful examples of obliquity, or perhaps better to say inversion, that I have in my life, is that I find my pessimism to be alleviated somewhat when I face the thing head on. I'm still powerless against it, but I find that when I engage with the topic, when I really read about the scenarios, even th... Show more

Claude skills are incredible, it's like a near term hack for solving continuous learning. But I think they are also unintentionally end up becoming a new and incredibly valuable source of training data for executing tasks with long time horizons. The current state of AI is such that it incentivizes you to train the... Show more

Hello restart my old friend