793 karma · joined May 9, 2018
The problem is it's hard to come up with better examples than your toy examples of weather and maps. Goodness there are so many travel planning demos. Who actually wants the context switch of a UI popping up mid-typed-sentence? Is a date picker really more convenient than typing "next school break"? Visualizations are interesting -- but that changes the framing from soliciting input to densifying information presentation. Datagrids and charts'll be valuable.
Anyway, it's a space that's still starving for great ideas. Good luck!
let's permit the firefighters to leave the firehouse even though they can't tend to all the fires simultaneously
This goes back to a principle of safety engineering: the safer, reliable, trustworthy you make the system, the more catastrophic the failures when they happen.
You’ve stated the widely-accepted formulation of Goodhart’s, but it can be interesting to note Charles Goodhart’s original statement was that “any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.” (The implication is people doubly go wrong presuming the regularity’s existence and placing pressure on it!)
> The programmer builds from pure thought-stuff: concepts and very flexible representations thereof. Because the medium is tractable, we expect few difficulties in implementation; hence our pervasive optimism. Because our ideas are faulty, we have bugs; hence our optimism is unjustified.
Humans just have bad intuitions for this problem space, so you have to be consciously empirical; externalize decision factors, track outcomes, articulate hypotheses and be honest.
And those duties can be achieved with today’s mechanics — they just need good control, which is now seeing ferocious progress
Imagine contrasting a house with the TV on all day vs one where kids have screen time with educational apps in the iPad.
Similarly, LLMs can be used as educational tools.
In the end, learning and self-improvement needs some non-trivial motivation from the individual.
The answer to your question is to show them it’s valuable to learn. If they find they’re completing their tasks adequately from AI assistance, then give them harder tasks. Meanwhile, model how your own learning effort is paying off.
Note how if they are left with AI-trivial tasks and the benefit of learning remains abstract, we shouldn’t expect anything to change.
Next, 1 essay, 100 fonts!…
I wonder if there are variants that have good baselines. It might be tough to evaluate vis a vis human performance on novel games..
The peril came from being led into dangerous waters. Maybe your analogy is more right than you know. The question is, who’s Titanic, Olympic, and Brittanic…
Thus we are left in this noisy, hype-addled discourse. I suspect these scientists are pushing against some perceived pathological thread of that discourse…without their particular context, I categorize it as more of this metaphysical noise.
Meanwhile, let’s keep chipping away at the next problem.
Team dysfunction is inseparable from tech dysfunction. Imagine one team owns three quarters of this service as another owns half of it. Or more likely, one owns a tenth of a problem space and another owns a fifth. Suddenly you need a fully realized solution. Who will step up? Well, you’re the one who actually needs it…but surely you don’t have to start from scratch? Surely…?
Then there’s just the fear of showing your ass. The day after it launches, who will be in hot water? Will it be you, because you agreed with the PM that your substandard code could ship? Why didn’t you flag the problem earlier? Get more help? Tell someone you weren’t already a superstar at this infra or that architecture?…
And then recall they did ship, it sucked, they bought Instagram. Then WhatsApp. And dozens of others
I guess the lesson is: yelling “go faster!” at a blocked-up engine is only gonna go so far. Mark’s (and many others’) move is to apply money to replace or get additional engines. Or roll up your sleeves and fix the org and culture…doesn’t have exactly “founder mode” romance to it tho, and there’s scarcely a harder task out there.
That’s it, folks. As always, the devil’s in the details, not the Startup School maxims.
As Ed Catmull said, (lightly paraphrased) “everyone says story is everything, even if their story is drek…once you reduce an idea to a concise phrase you can use it without risk of changing behavior. What really matters is, what are you going to do about it?”
Hint: We don’t spend our days scrolling horizontally
What’s respectful to users is a separate (but not wholly unrelated) question…