pull down to refresh
AOSP phone
Yes, I'm getting close to turning autocorrect off completely.
there will be new friction
This is something that it's very easy to overlook (as I did). There will be new friction just as there will be new jobs.
I find it interesting to think about cases where removing friction is a bad thing.
For instance, in the case of physical fitness, the friction is clearly good. I want the weights to be hard to lift. And I also think that I want it to be something I have to force myself to do. If it doesn't require a little willpower, it must not be hard enough. I don't want something to come along and make this easier for me.
I don't think this is quite how the article meant "the friction is the point," though. But maybe there is analogy: just as the friction in working out is essential to a healthy body, perhaps the friction in figuring out how to make something people want to use is essential to making things that are truly useful.
So, now, LLMs are sold as solving a lot of the friction of making something useful. But this may be a trick.
I'm afraid this sounds a lot like whining on my part. What is the use of such an observation? Well, it may be the same as advising writers to write with a pen on paper. Because something about the process is useful (not that a writer should only write with pen and paper, but that there is something useful about keeping it in the process).
Maybe it's the same reason I keep cleaning my own bathrooms. The wife has suggested that we ought to consider getting a maid to come by every once in a while because I complain mightily. But I think it's useful to clean my own messes. Also I'm a tightwad.
I find it interesting to think about cases where removing friction is a bad thing.
I had to think about this a bit more. I don't think we can ever fully remove friction. There are always tradeoffs. So we can reduce friction, and shift it to places where we don't mind it as much. But we cannot eliminate it; not really.
So, now, LLMs are sold as solving a lot of the friction of making something useful. But this may be a trick.
Remember Zuck saying "sharing is caring"? What he meant was: "sharing is good so that Cambridge Analytica can leech everything you ever shared straight from our DB and pitch you the Brexit narrative that will make you vote Leave and get filthy rich from that contract with the lobbyists."
Much of the AI narrative is the same trickery. They have to fake it until they make it because no one is going to just give their data. That's what I meant with lie. The reason why we know they lie is because Claude 3.7 was supposed to be able to do what Claude 4.6 still cannot really. They oversell. All of them are RalphWiggum-ing their pet projects spending millions on nonsense with mediocre results that then are oversold in round 2 of the marketing machine. "Claude developed a compiler"? Not really. Some dude spent 20,000 hours (of unspecified scale - H200-hours?) and then brute forced a compiler. That's not developing / engineering, it's called wasting valuable moneys.
Maybe it's the same reason I keep cleaning my own bathrooms.
It is, but not because you like the friction, but because you like the result! (I clean my own bathroom too, for the same reason, except a couple of years ago when I was recovering from a back injury.)
We could be, if it weren't a bit of a lie. Fortunately, I think that the lie is 2-pronged:
allcan only be limited in scope, e.g.all known friction, and that weakens the unpleasantness a bit, as there will be new friction.PS: are you on an AOSP phone? bot->not is the most awful thing it does to me, lol