Nothing wrong with wanting fairness against bullies. As a Canadian, I appreciate the support!

That said, I look forward to the day when the US is done with this insanity and we can go back to being awesome trading partners.

I think the core similarity is human communication trends (especially in English): sycophancy, prioritizing social status, the confident assertion of unclear claims, etc.

While obviously they all have a huge amount of overlap in their training data, I bet if you trained a model only on 19th century literature it would still pick up a lot of the same conversational tendencies (aside from maybe the bullet points) as current models.

As much as I was rooting for Grok 3 to hilariously suck, it does seem to do well on a lot of puzzles, including ones made up just to try to fool it. I'm confident they're about to be eclipsed by another lab though, for what it's worth.

I’m curious how this compares to the cost of developing the Internet (and migrating business to it). Most of AI costs go toward running AI applications (inference) and not specifically toward training new models. That said, the amount of investment shows a lot of faith in the tech.

I agree that would be nice, but it’s looking like a lot of the power of modern AI is from its ability to capably model the world – and this “world modeling” is what seems to give rise to self-awareness. I strongly suspect we can’t easily disentangle the two.

Threads was supposed to compete with Twitter but not outcompete their own Facebook and Instagram with a new more enjoyable, less ad-filled feed.

Wearing a suit in Silicon Valley? What are you, a Boomer?
(maybe good for super serious guests tho)