Matthew Prast

My problem with the "taste" discourse is that it's contrived. You would not single out taste as a uniquely human skill if you weren't scrambling to justify your own employment.

It seems almost certain that with enough observation a model could predict what you personally would like and what decisions you'll make (in fact some people were making exactly this claim 5-10 years ago - see, for example, StitchFix)

I've sometimes heard people say "oh you can't trust AI to get things right the first time, you have to ask it to correct itself" and that seems circular?? Like why would I trust it to get it right the second time if it can't do it right the first time

respectfully: 1) the overton window has been distorted by some truly extreme marketing re: AI's capabilities and future. views typically labeled as "anti-AI" range from "it's completely worthless" all the way to "LLMs will probably not be a suitable replacement for humans in every single case" /1

What the right is *definitely* interested in is consolidating money and power by making promises they can't keep and then keeping people just distracted enough that they don't catch wise. /3

respectfully: 1) the overton window has been distorted by some truly extreme marketing re: AI's capabilities and future. views typically labeled as "anti-AI" range from "it's completely worthless" all the way to "LLMs will probably not be a suitable replacement for humans in every single case" /1

the more I use AI the more I'm convinced the killer app is documentation. it's *ok* at writing code, but honestly I find myself relying on it less and less as the novelty wears off. when I'm using a language or framework I'm unfamiliar with and need someone to consult, it's incredible

when the dust settles I think this is where LLMs are going to shine. the agentic stuff...honestly not so sure

i guess i can go without tiktok (although i do get a lot of recipes from there, so that's kind of a bummer.) what bothers me more is that this ban is flimsy security theater that will do very little to improve the nation's dismal cybersec position

the more I use AI the more I'm convinced the killer app is documentation. it's *ok* at writing code, but honestly I find myself relying on it less and less as the novelty wears off. when I'm using a language or framework I'm unfamiliar with and need someone to consult, it's incredible

1) people think e*on is a genius, but 2) he bought twitter and drove it into the ground, which kind of contradicts 1) but 3) his twitter stakeholders are getting a huge payout anyway from his new AI company which raised a ton of money because 1) everyone thinks he's a genius

i know this is a joke but i feel like there's a palpable difference between companies that make money by making the customer happy and companies that make money by relentlessly optimizing their margins, and the second kind often charges more

the finished gingerbread! this is nigella lawson's recipe www.nigella.com/recipes/fres...

if you're in America and you'd like to try this, I highly encourage you to use golden syrup instead of substituting with corn syrup or agave - the taste really comes through

My Big Contrarian Opinion: one of the most damaging ideas in software development is the notion that you make an engineer more productive primarily by saving them keystrokes.

I would never abandon my beloved Brooks Adrenalines but anyone who wears them knows they are the best shoes that come in the WORST colors. I saw a pair that was pinkish tan and lime-green. Very excited to see the GT24s finally look like a cool modern running shoe

I don't think we'll disappear and I don't dispute that human intelligence is deep down probably nothing more than electrochemical interactions, but I do tend to be leery of reductionist perspectives like this because... 1/x

"oh the brain can be boiled down to lots of simple parts AND we know and understand all the parts and their interactions and will surely understand the brain in its entirety in 5-10 years." I think assuming the latter comes along with the former belies a certain overconfidence 3/x

I don't think we'll disappear and I don't dispute that human intelligence is deep down probably nothing more than electrochemical interactions, but I do tend to be leery of reductionist perspectives like this because... 1/x