For perspective, that's >$320k today. For 'switching apps'. It's always kinda crazy to look back and realize how incredibly minute and small programs were back them. Like 'spellcheck' being a project you'd assign half a dozen programmers to or Knuth would write almost a book about.
My recommendation is to have ChatGPT write code to solve calculations like this, and have it put the data it fetches as constants in said code.
AI is not a calculator, but it'll try to fake it - and above say, three or four digits, there's not enough training data to guarantee accuracy enough of the time. Labs probably train models on massive amounts of arithmetic, but still. It's not the right tool for the job.
Now, ChatGPT won't solve problems with code unless asked, but if you ask, it has its own full computer and can write code for it!
AI is powerful because it can use tools and can build tools for you to use or for it to use.
> AI is a powerful tool because it can use tools and can also build tools for you to use or for it to use. But it is not a calculator.
It’s both, and a million other things. Of course AI can do complex calculations today. I can’t even remember the last time I saw an AI make a mistake on something like that.
The problem is that people’s mental image of what AI can do always has an inertia of 6-12 months. Like you can still see people talking about how image generation models supposedly can’t draw hands, even though today we have 2K AI-generated video with perfect hands.
I've seen AI-generated videos posted two weeks ago with six fingers on someone's hand. I don't know what model they used, but it still happens. Less often than it used to, granted, but it still happens.
Reading these types of of stories growing up was always humbling.
I started at Microsoft on the Windows Mobile compiler team, having never taken a compiler class in college. I was surrounded by people who had already done great things, and by the time my career started the glory days of doing real software engineering was already waning.
I got to work with a few of the greats during my time as Microsoft, and I wish I'd taken up an in offer I had to meet Raymond Chen.
Sitting next to people of old school caliber and watching them work is, to use the proper word yet again, humbling as hell.
Fuck your design patterns. These people can debug starting at assembly, and sometimes when that fails they'll debug by pulling out an oscilloscope. Then when that is done they'll build design patterns on top of the foundations they laid themselves.
Multifinder could be thought of as a spiritual successor to Switcher but it shared no code with Switcher and took a different approach to solving the problem of allowing a user to interact with multiple running applications.
Erich Ringewald, the author of Multifinder, and Andy weren’t done with this problem with Apple. I got to work both of them, Erich at Be and Andy at General Magic/Eazel.
At least that’s what ChatGPT says
AI is not a calculator, but it'll try to fake it - and above say, three or four digits, there's not enough training data to guarantee accuracy enough of the time. Labs probably train models on massive amounts of arithmetic, but still. It's not the right tool for the job.
Now, ChatGPT won't solve problems with code unless asked, but if you ask, it has its own full computer and can write code for it!
AI is powerful because it can use tools and can build tools for you to use or for it to use.
But it is not a calculator.
It’s both, and a million other things. Of course AI can do complex calculations today. I can’t even remember the last time I saw an AI make a mistake on something like that.
The problem is that people’s mental image of what AI can do always has an inertia of 6-12 months. Like you can still see people talking about how image generation models supposedly can’t draw hands, even though today we have 2K AI-generated video with perfect hands.
Creator did you a favor making their slop so detectable ;)
Quick!
What's the next top-k tokens after each of these?
"999 + 999 = "
"999.0 + 999.0 = "
"999 * 2 = "
"999 * (3-2) = "
Mathematically identical. But as a token sequence, not so much.
Modern models will of course have very similar next-token distributions across all of these. But not identical!
Now add a few digits and you quickly land in territory whose synthetic training data is much sparser.
Code has determinism and runs in milliseconds. It's the better tool for the job.
I started at Microsoft on the Windows Mobile compiler team, having never taken a compiler class in college. I was surrounded by people who had already done great things, and by the time my career started the glory days of doing real software engineering was already waning.
I got to work with a few of the greats during my time as Microsoft, and I wish I'd taken up an in offer I had to meet Raymond Chen.
Sitting next to people of old school caliber and watching them work is, to use the proper word yet again, humbling as hell.
Fuck your design patterns. These people can debug starting at assembly, and sometimes when that fails they'll debug by pulling out an oscilloscope. Then when that is done they'll build design patterns on top of the foundations they laid themselves.
(The article doesn't mention this later became "Multifinder" and included with System 6 iirc.)
Erich Ringewald, the author of Multifinder, and Andy weren’t done with this problem with Apple. I got to work both of them, Erich at Be and Andy at General Magic/Eazel.
As a 1980s developer I wonder if that may have been a reference to Borland's Sidekick launched in 1984.
* https://en.wikipedia.org/wiki/Borland_Sidekick
Or was Memory Shift an actual TSR swapper I simply never encountered?
Switcher (2004) - https://news.ycombinator.com/item?id=32414661 - Aug 2022 (30 comments)
"Don't argue with me. $100,000 is fair, and you know it." - https://news.ycombinator.com/item?id=1637862 - Aug 2010 (110 comments)
Mac OS gets multi-tasking - https://news.ycombinator.com/item?id=1319308 - May 2010 (7 comments)