“Good writing is good editing.”

Also true with code.

I know Microsoft has the same credits-for-equity deal with OpenAI. I can’t figure out how that shows up in their financial statements. The ownership shows up as an asset on their books. But what’s the balancing transaction? Are they marking the credits as a loss at the average price-per-compute?

I once started putting together a D&D campaign that explored the impacts of hyper-inflation to a small village economy after an adventurer shows up and starts spending a massive amount of dungeon loot. Sadly, people weren’t interested.

I have failed this orbit during an embarrassing number of projects. The immense gravity of modern DevOps drags me beyond its event horizon until, hours later, my spaghettified brain simply gives up.

Source: www.nytimes.com/2024/02/09/o...

I think that axis is mid-labeled. Should be “Chill” at the bottom and “Not Chill” at the top.

Let me simplify that. When a flock of birds has lots of food, the ones who show off fancy but useless plumage might reproduce more. When food becomes scarce then simple, practical plumage helps you get your next meal. Suddenly, the peacocks can’t survive.

We put all the peacocks in charge.

“A chance to grow when the biggest tree in the jungle falls” is an interesting analogy. 10 years ago Twitter was talked about like the next permanent, unbreakable mountain of modern life. Now it’s a “fallen tree”. What we’re discovering is the transience of an industry we thought was more reliable.

> With just n = 8 they produce t(6) = 21.92, with a miniscule p-value.

I won’t say this never happens. But when it does it’s confirming something obvious and well-known. Like “ice cream melts faster when it’s hot out” obvious. And this is good. It should happen. A lot.

Also, can we all just agree that Excel does not leave enough of a log or paper trail to be used for any serious data work? It’s a great Swiss Army knife. But if you tell me you regularly perform surgery with that Swiss Army knife then I’m reporting you to the medical board.

As someone who has spent much of his career in finance…this tracks.

Whoa!!! This is really interesting!

“We find that use of model-generated content in training causes irreversible defects in the resulting models... We refer to this effect as model collapse… to avoid model collapse access to human-generated content is essential.”

I wonder how much is enough.

Legit question though. If the internet is flooded with AI-created content, and AI continues to retrain on that content, won’t it get kinda stuck spitting out the same sounding stuff? Like a copy-pasta echo chamber of early-2020s discourse frozen in algorithmic amber for future generations.