> Where's the one guy AI pioneer doing what used to take 100s?
Does me programming my phone into a magic wand from Harry Potter using a single spoken sentence count? Does me talking the phone into becoming a 3-way universal translator from Star Trek count?
In both cases, it's a trivial use of current (as of 6-12 months ago) capabilities of Gemini and ChatGPT, respectively, plus some basic logic. As simple as:
if(someone just said "Lumos") { togglePhoneFlashlight(); }
But in each case "implemented" totally spontaneously, on the fly, using less code than above (though in natural language), and it directly solved specific problems for real, concrete users (which is more that I can say for most coding work I did).
Is this much? Not really, but 5 years ago it would take at least a single engineer and at least a good work day (a conservative estimate) to properly wire all things together and put in a mobile form factor for testing. Today (minus 6-12 months) it took me less than a minute, so that's a 480x improvement right there.
And then I can tell you how last week I had Gemini single-shot me a proper in-browser photo editing tool for autocropping photos of scans (complete with auto-placing non-rectangular quad cropping zone for rotation and perspective correction, snapping sides to edges inside image for minute edits, and host of other trivial editing features). It's impressive it did that at all (even if all the heavy lifting was done by JS version of OpenCV), but it's not the important part. Neither is that result was correct on the first try. No, the important part - the thing that's profound about this technology - is that having AI build such a tool was much, much faster than finding an existing tool for that job.
Building tools to help yourself with your own work isn't a new thing (especially outside of computing). But LLMs are a qualitative change here - they let you wire up bespoke solutions in minutes, that would otherwise take hours or days of fully-focused expert work, which throws xkcd://1205 out the window, as suddenly it makes sense to do that to shave an hour off a two-hour task.
Does me programming my phone into a magic wand from Harry Potter using a single spoken sentence count? Does me talking the phone into becoming a 3-way universal translator from Star Trek count?
In both cases, it's a trivial use of current (as of 6-12 months ago) capabilities of Gemini and ChatGPT, respectively, plus some basic logic. As simple as:
or: But in each case "implemented" totally spontaneously, on the fly, using less code than above (though in natural language), and it directly solved specific problems for real, concrete users (which is more that I can say for most coding work I did).Is this much? Not really, but 5 years ago it would take at least a single engineer and at least a good work day (a conservative estimate) to properly wire all things together and put in a mobile form factor for testing. Today (minus 6-12 months) it took me less than a minute, so that's a 480x improvement right there.
And then I can tell you how last week I had Gemini single-shot me a proper in-browser photo editing tool for autocropping photos of scans (complete with auto-placing non-rectangular quad cropping zone for rotation and perspective correction, snapping sides to edges inside image for minute edits, and host of other trivial editing features). It's impressive it did that at all (even if all the heavy lifting was done by JS version of OpenCV), but it's not the important part. Neither is that result was correct on the first try. No, the important part - the thing that's profound about this technology - is that having AI build such a tool was much, much faster than finding an existing tool for that job.
Building tools to help yourself with your own work isn't a new thing (especially outside of computing). But LLMs are a qualitative change here - they let you wire up bespoke solutions in minutes, that would otherwise take hours or days of fully-focused expert work, which throws xkcd://1205 out the window, as suddenly it makes sense to do that to shave an hour off a two-hour task.