Everything Alex Graveley said on any show that made the record, most notable first. Each card names its show and opens the statement there.
Graveley: Bad actors with AI pose a bigger threat than AI existential risk
“I'm not really scared about AI is killing us, although I'm very grateful that there are people that are thinking about it. I'm more worried about bad people using new technology to hurt us.”
Training OpenAI base models on code improved chain-of-thought reasoning
“So they had wanted to see what introducing code into their into their base models would do. I think it had positive effects on chain of thought reasoning.”
Graveley: Every 10ms of latency drops Copilot completions by 1%
“Every 10 milliseconds is one percent fewer completions that people would do.”
Graveley: Programmers think in blocks, making block autocomplete optimal
“The key insight was realizing that programmers kind of think in blocks. So if you can just autocomplete the rest of the current block, Then that's what people want. People can, they're kind of like judging what this block is going to do, and if it does the rig…”
Graveley: Copilot writes 80% of code for top power users
“There's some whales out there. Like there's people who it writes 80% of their code.”
Autonomous tax-filing AI agents are achievable within the next few years
“You want to make a thing where you say, hey, computer, file my taxes, and it does the right thing. You know, I think we can get there. That's certainly in the next few years you know.”
Graveley: Applied AI intuition is bimodal between junior and senior engineers
“You know, one thing, I don't know if other people have noticed this, but it seems very bimodal, like very senior people get it and very young people get it, and so I don't know that the middle has really caught up yet or realizes that, that .”
Graveley: Unlike most software that breaks at scale, AI models improve
“The crazy thing is that these things work at all, right? And not only that, but they scale and they improve the scale, right? Most things break at scale. Almost everything breaks at scale.”
Graveley: Traditional UIs enforce structured flows, whereas text interfaces invite erratic goal changes
“So the beauty of the web and like traditional UIs, right, is that you fill them in. And like, if it, if you can't, if it doesn't do what you want, then you make the decision to either go forward or not go forward. Right? Text has this annoying property of you …”
Graveley: AI scamming will cause years of extreme discomfort and financial harm
“The honest answer is I think we're going to go through a many year period of extreme discomfort where AIs pretend to be things or confuse people or extract money from your grandparents or you know, drain people's life savings in ways that are that are scary”
Graveley: GitHub Copilot's in-the-wild test pass rate rose from under 10% to over 60%
“And eventually, you know, it went from, you know, less than 10% on in the wild test to over 60%. So that's like one in two tests it can just generate code for.”
Graveley: GitHub Copilot weekly cohort retention stayed above 50%
“Our retention rate was 50%. Like, it never went, like, months later, it was still above 50% by, like, weekly cohort, which is, like, insane.”
Web interaction click data is a massive and unowned dataset
“So people clicking on stuff on the web is a gigantic data set, which is currently unowned.”
GitHub initially considered building a Stack Overflow competitor using OpenAI models
“And at the time they were thinking in terms of like, maybe you can replace Stack Overflow or something, you know, do a Stack Overflow competitor.”
Graveley: Early data centers and networks ran primarily on Windows
“Windows was popular at that time. So people ran their, you know, data centers on windows and networks on windows. And this Linux thing was like very strange and people didn't really know what it was.”
No Priors 2