Less wrong

Average, or worst?

December 30, 2022

Over the last few years, deep-learning-based AI has progressed extremely rapidly in fields like natural language processing and image generation. However, self-driving cars seem stuck in perpetual beta mode, and aggressive predictions there have repeatedly been disappointing. Google’s self-driving project started four years before AlexNet kicked off the deep learning revolution, and it still isn’t deployed at large scale, thirteen years later. Why are these fields getting such different results?

~ Alyssa Vance from, https://www.lesswrong.com/posts/28zsuPaJpKAGSX4zq/humans-are-very-reliable-agents

slip:4ulepo2.

This makes the interesting distinction between average–case performance, and worst–case performance. People are really good by both measures (click through to see what that means via Fermi approximations.) AI (true AI, autonomous driving systems, language models like GPT-3, etc.) is getting really good on average cases. But it’s the worst–case situations where humans perform reasonably well… and current AI fails spectacularly.

ɕ

Cole’s law

July 16, 2021

Hofstadter’s Law – “It always takes longer than you expect, even when you take into account Hofstadter’s Law.”

~ “rogersbacon” from, https://www.lesswrong.com/posts/Mt3vtAQGnkA3hY5Ga/eponymous-laws-part-3-miscellaneous

slip:4ulepo9.

It’s part 3, and it is a nifty collection of serious and whimsical laws. However, I doubt that Stigler is the originator of Stigler’s Law. Sometimes the only reason I write this stuff is to see if I can entice you to go read the thing to which I’ve linked.

But more often I do have a point. I’m wondering, in this case, how much of our urge to create, and our delight in such pithy Laws as Dilbert’s, comes simply from our mind’s desire to find patterns. There are a slew of cognitive biases, (confirmation bias springs to mind as fitting the pattern of my example,) which feel like they arise from pattern matching gone overly Pac Man.

ɕ

Math

May 14, 2021

Predicting the behaviour of a sigmoid-like process is not fitting the parameters of a logistic curve. Instead, it’s trying to estimate the strength of the dampening term – a term that might be actually invisible in the initial data.

~ Stuart Armstrong from, https://www.lesswrong.com/posts/6tErqpd2tDcpiBrX9/why-sigmoids-are-so-hard-to-predict

slip:4ulepo4.

Wait! Don’t flee!

It’s a great explanation of sigmoids—you know what those are, but you [probably] didn’t know they have a general name. People toss up sigmoid curves as explanations and evidence all. the. time.

Ever make that slightly squinting face? The one where you turn your head slightly to one side and look dubiously, literally askance at someone? …that face that says, “you keep using that word, but I do not think it means what you think it means.” After you read that little article about sigmoids, you’re going to make that face every time some talking-head tosses up a sigmoid as evidence for a prediction.

ɕ

The answer is, “2.”

November 26, 2020

In this situation, before committing to a three year PhD, you better make sure you spend three months trying out research in an internship. And before that, it seems a wise use of your time to allocate three days to try out research on your own. And you better spend three minutes beforehand thinking about whether you like research.

~ ‘jsevillamol’ from, https://www.lesswrong.com/posts/eZCrCB3HiDB55Ccqx/spend-twice-as-much-effort-every-time-you-attempt-to-solve-a

slip:4ulepo11.

This one caught my eye because the vague heuristic of spending increasing amounts of effort at each attempt to solve a problem felt true. But I was thinking of it from the point of view of fixing some process— Like a broken software system that occasionally catches fire. Putting the fire out is trivial, but the second time I start trying to prevent that little fire. The third time I find I’m more curious as to why does it catch fire, and why didn’t my first fix make a difference. The fourth time I’m taking off the kid gloves and bringing in industrial lighting, and power tools. The fifth time I’m roping in mathematicians and textbooks and wondering if I’m trying to solve the Halting Problem.

Turns out the context of the problem doesn’t matter. The answer is, “2.” Every time you attempt to solve a problem—any sort of problem, any context, any challenge, any unknown—the most efficient application of your effort is to expend just a bit less than twice the effort of your last attempt.

Not, “it feels like twice would be good,” but rather: Doubling your efforts each time is literally the best course of action.

…and now that I’ve written this. My brain dredges up the Exponential Backoff algorithm. That’s been packed in the back of my brain for 30 years. I’ve always known that was the chosen solution to a very hard problem. (“Hard,” as in proven to be impossible to solve generally, so one needs a heuristic and some hope.) They didn’t just pick that algorithm; Turns out it’s the actual best solution.

ɕ

Meaningful rest

October 27, 2020

Worse, this combines really badly with my default working style. I have a strong neurotic desire to finish things, and to fixate on my total output rather than time spent working. I’ll often push myself to complete my current task, going well beyond my allocated working time, and not being willing to take a break until I’m done.

~ From https://www.lesswrong.com/posts/iYR9hKzTKGhZwTPWK/meaningful-rest

slip:4ulepo12.

This is an article examining what it means to rest, and how breaking one’s default behavior is critical. The bit I’ve quoted was definitely a problem for me. Changing my default thinking in the form of “shoulding” on myself has opened up several other doors to change. (Note that I still, very carefully grammared around implying I’ve been successful at change—that’s a default I’m still working on. :)

What’s your default that’s holding you back?

ɕ

Bad planning

August 24, 2020

It’s a long to-do list that doesn’t translate into action. A spreadsheet where you gather information in order to forget about it. A long chain of thought culminating in an epiphany that goes nowhere. An argument about an issue that you never work on directly.

Bad planning like a belief in telepathy. It makes you feel like your private thoughts can change the world. The quintessential example? A college humanities essay that gets read by the student, the professor, and nobody else, but which the student remains proud of for the rest of their life.

~ “AllAmericanBreakfast” from, https://www.lesswrong.com/posts/pt7XYxnhTQdHGf7MN/change-the-world-a-little-bit

slip:4ulepo15.

Planning, todo list management, goal setting… for me it all comes down to beginning with the end. What does “done” look like? What would a solution to this problem look like if I had a magic wand? When this is done what effect will it have [on me or the world]? Far too many people struggle with lists, and with getting things done—also with Getting Things Done. The real challenge is to figure out if the idea you just had pop into your head… is that a how to do something, or a what [as in, a goal] to do? If you have a how you really need to figure out that what. Because otherwise…

How are you going to figure out why you are doing anything that you are doing?

ɕ

Quiet

June 19, 2020

I maintained this illusion until, inspired by a stupidly expensive device that only does one thing, I taped my old phone to a bluetooth keyboard and began to write in offline mode. It was immediately a magical experience. It was so *quiet*. I could go on my porch and write and it was quiet. My thoughts got much larger because I wasn’t subconsciously afraid I’d interrupt them. I began to feel angry at my laptop. Why did it insist on hurting me so much? Why couldn’t it be pure like the offline phone/keyboard experience? Why couldn’t I just create things?

~ “Elizabeth” from, https://www.lesswrong.com/posts/To5xcWvjN744rWEKQ/turns-out-interruptions-are-bad-who-knew

slip:4ulepo3.

Don’t worry, I’m not getting on my soapbox about distraction and being used by your phone and the Internet and social networks. Nope, definitely not getting on my soapbox.

Today, I’ve gotten my ladder and I’m climbing on my roof to preach right over your head—you nice people in my front lawn, who are smart enough to be reading, reading stuff that has paragraphs, from a site on the open web, even if you only subscribe to the email because you haven’t mastered RSS—nope not preaching at you, dearest choir of mine, not today.

But you people in the back… Can you not see the Oxo® easy-grip handles that extendeth from thine brains?! Can you not see the unwashed masses of people who labor for Facelessco et al to write software that grabs you by those handles?

What say you? WHAT? …sorry you have to yell, I can’t hear you so well from up here on my roof… Oh, you cannot in fact see the handles? …well, have you tried looking in the mirror? …uh, hello?! Where are you going? Oh yes, definitely check that message, and scroll through Instagram and I’ll just wait here on my roof.

ɕ

Variolation

April 4, 2020

I generally don’t write about current events here on my blog. But occasionally I find something that I think would be so beneficial for more people to read, that I find I want to share it.

Initial viral load seems likely to have a large impact on severity of Covid-19 infection. If we believe this, we should take this seriously, and evaluate both general policy and personal behavior differently in light of this information. We should also do our best to confirm or deny this hypothesis as soon as possible.

From, https://www.lesswrong.com/posts/3ArEA7tHDXQxE6PED/taking-initial-viral-load-seriously

slip:4ulepo5.

Since virology has taken such a place of primacy in our lives for the foreseeable future, it can only benefit each of to read more. That article is a wide-ranging, opinion piece (so, I recommend a few grains of salt with it,) which touches on a treasure trove of topics and facts. Of particular note is its discussion of how vaccines work for other diseases. (Or maybe I should write, “…of the variation in efficacy of vaccines for other diseases.”)

ɕ