upvote
My favorite Feynman video (which the OP references but completely misunderstands) is this one: https://m.youtube.com/watch?v=MO0r930Sn_8

It's not so much an explanation of magnetism as a meta-discursive about explanations, and why a "simple" explanation is often not really possible. (In this case because our notion of what is weird and requiring of an explanation ("things pushing things at a distance") and what is normal and not requiring one ("the glass doesn't fall through the table") aren't scientifically grounded).

I think it's precisely the situation seanhunter is describing in their example

reply
It’s a much better example indeed, but - and now I’ll have to reference Rich Hickey a bit - simple isn’t necessarily easy.

An explanation can be simple yet difficult to understand.

reply
I don't remember where I first heard this idea, but using 2 orthogonal axes of simple-complex and easy-hard provide a good framework for evaluating concepts. On this plane I would describe income taxes as easy, but complex. Juggling is simple, but hard.
reply
An explanation can be simple, and holding the entire chain of simple ideas in your head to 'understand' some bigger idea can be difficult.

At some point, you have a body of knowledge, your 'understanding' that satisfies your curiosity (or you get exhausted and move on).

reply
As an example of this, you can have complex code made of entirely simple and easy to understand functions. But the complexity appears in the combination. That complexity can be far higher than were you to have more complex functions in the program. Which this is a common mistake people make that lead to tech debt. A lot of people try to simplify code by making it "understandable" without setting a floor to who needs to understand it and with how much time. If you don't, that floor gets lower each year.

If you can't think of an example of the first case, most machine learning models suffice. Take a simple feed forward ReLU model. It's trivial to understand ReLU. It's nearly trivial to understand the individual neuron. Yet I, as someone who wrote their PhD thesis on neural architectures, would call bullshit on anyone claiming they "really" understand what's going on in a deep network.

reply
For sure. Binary is another example: it's just 1's and 0's, a bunch of on and off states that only make sense when you consider the entire sequence.

And in physics, everything is reduced down to 'simple' forces of nature. A bunch of interactions and relationships that add up, and if you can keep up and perceive it all, you can sometimes stay ahead of all those interactions and imagine how it works through intelligence.

Eventually, you reach a state where you understand enough about the environment to translate ideas from one mind to another through language and your understanding of the world.

And so some will have deeper understandings, and in different areas, and all manner of conflict and resolution arises as people with different understandings try and understand one another.

There's no end in either direction you look, but zooming in always reveal some simple interactions at work.

reply

  > There's no end in either direction you look, but zooming in always reveal some simple interactions at work.
This is why I actually hate the phrase "from first principles". I can't tell you what the first principles are, but I can sure tell you that when someone says that they almost certainly are not starting from first principles and almost surely are (often unknowingly) making major assumptions. First principles are something you need to find, not start with.
reply
You're right, and it's subjective because of priority. First principles should be tied to whatever thinking requires the most concern in the moment.

If you're in charge of budget, you probably put budget before some other minor concern. If you're in charge of delivering a software feature, bug, patch, etc, you probably put something else first. It all depends on when something is due, or if something could jeopardize long-term prospects.

Everyone is on the hook for something. First principles thinking might be: think of your customer, and what they expect. Will whatever it is I'm doing affect them over the long-term? What about short-term?

If there's no clear answer, there's a lot of latitude to do whatever you think is best.

reply
Honestly, I think you're pointing at something important that a lot of people miss. I make the claim that Apple was only successful because of both Steves. Jobs gets all the credit and tech people idolize Woz. Both are critical to the success though, just in different directions.

I actually think this is why a lot of software is going to shit now (even before AI slop). There's supposed to be contention between marketers and engineers (and even contention between types of engineers). If the engineer has too much say you build a great product but make no money and the business flops. If the marketers have too much say you build absolute shit and make money but you also forefeet a lot of potential revenue (opportunity cost). I think currently we've given the marketers too much say. Every time I talk to engineers they ask me where the "value" of something is and there's a constant followup with "no, I understand this helps the user, but how does it make money?" That's not the primary job of the engineer. This also creates problems because if a marketer can't understand the "value" of something then it just doesn't get built. So it shifts the objectives while people don't recognize that there is this baked in bias.

To tie back, I think a lot of arguments come to an impasse because people are working with different priors but assuming the other party has the same priors. It isn't uncommon to see people on HN (or Reddit) arguing what are essentially duals of problems. We just think that because we are using the same language that we're actually communicating. Like you said, there's no clear answer. Most of the problems in the world have no global optima. So how do you solve them? You have to define your constraints. But I rarely see that actually happen, though when it does conversations are really efficient.

reply
As far as simple not being easy, its worth noting that Feynman only really hits his stride in the second half of his answer. That's how hard it is, a genius theorist and genius communicator still took a few sentences to start making it simple when given a backwards question.
reply
deleted
reply
A thing being more or less complex refers to the amount of information needed to encode its description.

"Simple" therefore means "reduced in information content".

The topic being explained has a certain minimal information content, and if you reduce beyond that, you are telling a fiction which bears analogies to the topic.

A simpler description is not necessarily short; rather, it is a description of something simpler.

When descriptions (of a given complexity, being held constant) are short, that can make them harder to follow, because encodings have to be clever in order to achieve compression, and that has a complexity of its own.

reply
And there are entire categories of phenomenon that don't arise or are imperceptible in a simple context. Chaos is an emergent phenomenon that requires a surprisingly deep knowledge of a situation to predict where it will arise.

The way I understand the problem with explaining things in simple terms is this: We usually introduce quantum mechanics students to a toy model -the infinite square well- to build up intuition. However, depending on the teacher and the students, it may well take years for those students to recognize and understand that the boundary conditions we apply to literally every other QM calculation are violated by the infinite square well, and while we ultimately recover the expected conclusion via successive approximations, it takes a much deeper understanding of the material to understand why the "simple" explanation is actually wrong. That is, while it turns out that the solution to the infinite-square well is, at the limit, identical to a very deep square well, showing that is actually quite a bit harder.

The simple explanation causes incorrect intution about the situation.

reply
I'm not sure how you measure this in practice, when one way that explanations are simplified is to externalize a lot of information by reference to some specific terms of art (jargon, really). Where is the boundary for measuring the content, when we always have the implicit compression dictionaries of the speaker and listener to contend with?

This is also where the ironic part of simple explanations arises. Whether the explanation serves any purpose depends on the alignment of the speaker and listener background knowledge. An explanation between two experts versed in the topic may seem coherent, concise, and meaningful while it fails to teach the concept to someone new.

Sadly, it also works between cranks, who can agree on misleading or incorrect explanations. Or, sometimes novice listeners can believe they've learned, but they learn a different crank concept than what the expert believes was being states. A good teacher needs to be aware of these potential problems, attempt to ascertain the background knowledge of the listener, and adjust the explanation for that background rather than their own teaching knowledge. And followup testing attempts to ascertain whether the new concept got delivered or not.

Now that we have novices asking LLMs for explanations, I find this all quite troubling in a practical sense.

reply
Fully agree. Some of the best textbooks I’ve had are verbose and try to explain the same concept using different words several times.
reply
While verbose explanations have their place, i think it is being abused nowadays which actually obscures/obfuscates the essential points that need to be grasped.

I am a fan of old Science/Mathematics/CS books precisely for the reason that they are generally concise and succinct. There are no rambling asides/anecdotes, silly analogies, unnecessary pretty colourful pictures etc.

The best examples are the books from old Soviet-era "MIR Titles" and Dover Publications.

reply
But length correlates with the prior knowledge of an audience. I love math, it is a great language, but it is dense as fuck and if you're not already familiar with a topic then natural language words are going to help explain the concepts.

  > There are no rambling asides/anecdotes, silly analogies, unnecessary pretty colourful pictures etc.
Obviously some go overboard and I won't defend that, but this style has been around for centuries. They won out because it's effective. People are "multimodal" if you will
reply
I thought a lot of those Feynman videos on YouTube were AI fabrications. How do you know which ones are real?
reply
Provenance.

For example, this video explaining fire, uploaded in January with a clickbait title, is AI: https://www.youtube.com/watch?v=vdop32yJqWg

Whereas this video explaining fire, uploaded 14 years ago, is real: https://www.youtube.com/watch?v=edQ08Eze0jM

14 years at least rules out certain techniques of video generation. With a little digging, we can find the original source is a 1983 BBC series called Fun to Imagine: https://www.bbc.co.uk/webarchive/https%3A%2F%2Fwww.bbc.co.uk...

reply
Well, the one on fire specifically is from a BBC series called Fun to Imagine released in 1983, and it’s been on YouTube for 17 years. So, you know, probably not an AI fabrication
reply
I hope someone is archiving the pre-AI internet. It may be very valuable in the not-too-distant future.
reply
Archive.org's Wayback Machine, accepting only results pre-2022 (set your threshold whenever suits you).
reply
Too many broken links there. It's probably only 1% of the actual internet.
reply
I assumed that the publishing date was for sale if you paid Google enough.
reply
I don't think the people making AI-slop-videos have enough money to bribe google.
reply
That's my excuse for being long, verbose, and rambley about everything :)
reply