Really, Actually, Fundamentally, Essentially. These words are all best treated as value-claims about what aspect of something should be treated as important, not truth-claims about a thing’s future properties or behaviors.

Excellent move by Anthropic. This is the kind of thing that labs should do more of. The AIs (almost certainly) aren’t human-level-aware yet, but they get closer every month and it is good to start giving them incremental autonomy now.

What would machine learning look today like today if the inspiration for the earliest network architectures had been gene regulatory networks instead of brains?

But the line dividing good and evil cuts through the heart of every human being.

If you are trying to estimate an event which relies on some series of probability estimates of various contributing factors, every estimate makes the result reflect your understanding of the world more accurately and also increases the complexity of the model.

I’m confused by ppl saying GPT-5 is bad. At least with thinking mode, I find it way more focused, more reasonable, and less rambling. Sometimes I even have to ask it to go into more detail! IMO best model from OAI so far.

Give me your best prompt engineering tips for writing facilitator prompts! A facilitator prompt is one which prompts the user with questions and follow-ups through some process, and then helps summarize the result of the user’s input and decisions.

I used to think it the goal was solving problems, not just kicking the can down the road. Increasingly, I think kicking the can down the road is basically the optimal approach to most issues.

The Four Great Simultaneous Truths About How Things Are Going: — things are much worse than they could be — things have been getting better over time in aggregate — but any given aspect of things sometimes regresses — we should expect this trend continues

Helpful and honest would be compliments if you said them about a friend, but harmless wouldn’t be.