jackrust.bsky.social

Slightly spicier: I say this as a guy who writes jokes and non-fiction... I think getting our politics via comedians is generally a bad thing. Getting jokes about politics from comedians? Great. But not our politics. Comedians can act as pundits, but then later say "I'm just a comedian."

I'm still annoyed that the OpenAI marketing campaign to investors that "we're building something that will probably enslave humanity" actually worked.

That shouldn't work. Surely that's both obviously not true and also bad.

Because LLMs *are* cool. Really cool. And can do legitimately interesting and useful things.

But because they pass the Turing Test and are marketed as forerunners of a vengeful AI god people forget that they are just probabilistically querying the corpus.

I wish LLMs hadn't taken off so I could just bore my wife with how cool they are (as with word2vec, Markov chains, and artificial neural networks before them)

Instead I'm always having to talk them down to correct the overhype.

So what am I supposed to do here? Aus-M and Aus-W playing tests at the same time.

My boss says that I should be working which I don't feel really addresses the core problem.

#SLvAUS #AUSvENG #TheAshes

When you watch him in the form he's in now, you remember why Smith averages 56
#SLvAUS

Like, the fact they can train comparable models for much cheaper, with lower quality chips is genuinely cool.

But it's not some random guy, it's a well funded start-up.
They're not starting from scratch, they're building off OpenAI responses.

I'm enjoying the DeepSeek drama but, like so much AI reporting, the actual cool bits are getting buried under piles of marketing and myth making.

Opinion | AI Can Work for Us A.I. can be designed to dramatically enhance individual empowerment.

This is such a weird article. None of Reid's ideas for empowering AI are particularly interesting or useful.

It doesn't read like an article about AI at all, just a vague defence of the concept of sharing your personal data.