Ever think of restarting the War Stories podcast?

I would also suggest Sam Waterston as a consensus pick.

My over-under on this episode is 12 hours

(seriously, this model is OpenAI's cyber truck - an embarrassing failure from what had been a respectable brand, but hyped so hard they can't walk it back)

Yeah that all checks out. Even get why @drewmagary.bsky.social didn't write it himself - we're so bad we don't even deserve another of these until we get our shit together.

And... now it's paywalled AND not even written by Drew. Well, f me.

The Patriots one is always excellent too (though these days, just kinda sad)

also, using it for what is good at rather than what people want it to do but isn't good at

What I'd really be curious to know as a follow-up: how much the performance of fine-tuned models varies on each? That's an area where I could see substantial differences existing between providers that aren't obvious until you've tried them all.

Cache insertion? Believe it or not, also search.

(a "vodka under the bridge" joke would've been more satisfying, though)

But isn't that one all just sand under the bridge at this point?

(not sure what that means either, but it's been a long day)

My curiosity was mostly whether there are ai assistants that, given a prompt to build an integration package, would start by looking for open code that already does that and then reusing it without attribution. Like, could it be an innocent(-ish) mistake?

Curious whether it was "IBM" or one underperforming engineer who thought they could get away with it.

(Alternatively, I'm curious if any coding assistant tools think it's fair game to copy and paste any code with an attribution-free license verbatim.)