I think this idea is often misunderstood and used as an excuse by people who don’t want to make any effort to understand a complex concept and want to blame the results of their laziness on the explanation or the explainer.
There are things where you may be given an explanation in simple terms, but the concept requires effort on the part of the recipient of the explanation to really understand. I’ll give an example: compactness. Probably anyone who has done analysis is familiar with how this concept is usually taught. Typically how it goes is you do open and closed sets and then they hit you with “a set is compact if and only if every open cover has a finite subcover”. Now this definition is extremely profound but very dense and it’s highly non-obvious why that in particular is the property you want (a finite subcover for every open cover)[1], how you could go about proving that every possible open cover in a certain situation could have a finite subcover etc etc.
In this case it’s actually quite easy to give simple explanations of the separate parts of the definition (open and closed sets, what an open cover therefore means, what a finite subcover then is etc) but actually understanding the concept of compactness requires something above and beyond whatever explanation you are given: it requires the reader to get out pencil and paper, try out some sets, go for some walks, think about it in the bath etc.
You may get a really fantastic explanation in simple terms from someone who deeply understands the topic, but if you don’t put in the effort you’re not going to understand it yourself. That’s not the fault of the explanation. It’s not an easy concept to grasp so you have to climb the ladder of understanding. The explainer can’t do that for you.
[1] it took the mathematical community years to settle on this being the definition, so you shouldn’t expect to understand why this is the definition straight away.
The article is just a bunch of quotes from Richard Feynman and John Gruber. There is little insight and I even think the article is misrepresenting the quotes.
Feynman said "if you can't explain it on a freshmen level". Leaving out that qualifier distorts the statement.
The other quote about the magnets is also overly reductive. Saying "I don't understand it in terms you're more familiar with" is not an admission that scientists as a group don't understand it at all.
You could go into what it means to understand something, but unfortunately the article isn't doing that.
Besides, the claim on magnetic force already serves as a counter-example to the main claim of the article.
Unless the article means to say that Feynman didn't understand the magnetic force, they have to admit that the magnetic force was perfectly understood, but impossible to explain in simple terms.
Explain to what level? E.g some topics common here. How do you explain NP in simple terms? "Things that are hard to compute?" That's simple, but not much of an explanation. It leaves the listener with a practically useless definition. And how do you explain even something practical as Heartbleed or Rowhammer? Etc.
And then there's another barrier, of course: the audience. You may use simple terms, and tell a lengthy story, but some things are hard to understand. Relativity and Monty Hall come to mind. They can be explained in simple terms, but how many people get them? Probably just the ones that also understand a detailed explanation.
So, should you also understand your audience, too? Is that ever possible? What if there's 300 people who listen to you with their own worldviews, viewpoints, health conditions (see different colors), life experience?
Literally yes, as much as possible. When preparing for an audience, you generally want to consider who your audience is when you're preparing your communication.
It doesn't matter what you know or understand, if you show up to a Spanish class and give a lecture in French.
I don't think explaining NP is that bad. Most people can understand what an algorithm is, some basic problems like sorting a list or TSP. If you just want to distinguish polynomial-time from non-polynomial-time algorithms, it's straightforward to give "obviously bad algorithms" (like brute-force search) and "obviously better algorithms" (pick your favourite) and to make that intuitive distinction precise.
Practical things like computer vulnerabilities are pretty easy to explain. I think anyone I know could understand what a buffer overflow is and why it's bad. The computer is a dumb machine and clobbers whatever you tell it to.
Relativity and monty hall are much better examples, I think. Relativity is well-studied enough that the explanations available for teenagers are pretty good, and if you study them you can get a pretty comprehensive understanding. But you do have to study them. Both are vulnerable to a certain kind of passive understanding, where correct arguments and incorrect arguments can both sound equally convincing.
E.g. NP is a problem that's easy to check, but not easy to solve. NP is type of a problem like one for finding people: it's easy to confirm (is Mr. X in the room?), but there's no quick way to find solution (where is Mr. X).
Lets put it this way - you shouldnt be a teacher probably if you can't even wrap your head around this while everybody else seemingly can.
I have yet to meet a single problem that when broken down to atomic parts wasn't trivial to grok. Every single piece of software. I am not saying 3 year old should understand details of general relativity, thats not the point.
> I have yet to meet a single problem that when broken down to atomic parts wasn't trivial to grok.
For one thing, we don't know how to break every problem down to "atomic parts". More seriously, the important things to grok in life are the things that are not atomic parts! If you look at quantum mechanics as a system of fancy mathematics with arbitrary rules, it is possible to become an unthinking machine that calculates outcomes. But that is not at all what it means to "grok" quantum mechanics! Likewise with relativity - it is very easy to have Lorenz transformations memorised but be totally unable to explain how e.g. the twin paradox works. Saying that's trivial to grok is utter BS.
It seems like you don't understand those things. I can envision explaining them in simple terms. NP isn't very complicated. xkcd had a great Heartbleed explanation in visual form https://xkcd.com/1354/
I don’t think this is true because it makes an assumption that reductionism is always possible, you can always distill every increasingly complexity into something less and less complex and maintain the overall behavior. I don’t think complexity is always reducible because sometimes the detailed nuance is absolutely critical to the distinction.
You may be able to take core ideas from something and distill them but when you have multiple interacting core ideas you may not necessarily be able to reduce them to simple terms. And almost always when you do there’s a big asterisk that the analogy has a lot of flaws.
I think it's not really about making simple analogies, because they don't really "explain" anything in the proper sense of the word. An analogy might make somebody feel like they understand it, but it doesn't really give actual understanding (see pop sci explanations of quantum mechanics).
It's more about breaking down stuff into simpler parts. For example, you want to explain a buffer overflow exploit in a web server. Then you'll ask your audience, do you understand addressing / memory allocation, and do you understand HTTP? If they say no to any of those, then you'll have to explain e.g. memory allocation first. And if they don't know how a CPU / RAM / caches operates, then you'll have to explain that first. And maybe you'll have to explain boolean algebra and how to build an adder from logic gates. And so on.
So I would claim that "reductionism is always possible". More complex things just take longer to distill into less complex things, but it can always be done. The exception would be things that "nobody really understands", like Feynman says about magnetism:
> I really can’t do a good job, any job, of explaining magnetic force in terms of something else you’re more familiar with, because I don’t understand it in terms of anything else you’re more familiar with.
Well I would push back on that claim because I don’t feel like you’re reducing things. You’re filling in gaps in understanding of something they already understand across the entire conceptual landscape it’s just masked by prior knowledge, they already know other concepts in detail and using a strategy to break a larger problem into smaller parts to make it digestible over time, you’re still dealing with the full set of complexity.
I think reductionism absolutely is possible in a lot of cases but I tend to agree with Stephen Wolfram’s concept of “computational irreducibility” or concepts we essentially know from compression in information theory (like Shannon entropy, Kolmogorov complexity) that there are limits to just how much you can reduce an idea or concept, and the bound for the human mind is likely much higher than those lower limit concepts.
Magnetic force is pointed out because it’s a more unique experience people don’t have a lot of prior knowledge and experience baked in that you can leverage as an assumption to reduce the concept, I think this holds true with lots of things. That isn’t to say we shouldn’t try to explain things simpler, just that that big caveat asterisk becomes ever more important when we use reductionist explanations. It’s not unlike when people try to explain spacetime curvature with a ball on a rubber sheet and things rotating around it… sure, that kind of gets at the idea, but there’s a huge asterisk that it’s simply not what’s really going on, it hints at the concept of curvature on a 2D plane you need to somehow conceptually extrapolate to other dimensions.
CS Lewis, an academic in a completely different field (literature: Professor of Medieval and Renaissance Literature at Oxford), said something very similar.
Being able to break things down into simple explanations (maybe many of them though?) is a sign of rigorous thinking. Not being able to do so (which seems common in some fields) is a sign of the opposite.
Yes, this is often obscured when working with code because it's relatively easy to judge its correctness. When using Claude to write documents, describe a system or organize requirements it becomes clear very quickly that Claude doesn't really understand what it is writing, it just puts probable things close together.
I think the problem here is the coding agent that we are working with are not familiar with the degree of knowledge we have around the system
It's like throw a lot of concepts at once that were just created 5mins earlier but the agent doesn't care that the human supervisor isn't aware about them and how they function, so it doesn't know what simple explanation in relative terms should be provided next
I think this is also implied in the article in the ending paragraphs
When making such strong claims that could easily devolve into semantics, it would be valuable to explain what you understand to be the definition of the semantic fulcrum of the claim.
In this case, that semantic fulcrum is what it means to '''understand''' something. Because the question of whether Claude understands things hinges much more on the criteria for understanding, than it does on our belief on the capabilities that Claude has.
i think a big part of my non-progrmamming IT work is pedagogy: many people build systems (even before AI) without being able to explain why/how they work
As everything, I think it depends. When the topic under discussion is complex and requires extensive background knowledge, it’s still possible to explain it in simple terms but the risk is to oversimplify it and loose important details.
The alternative would be to analyse every sub-topic until the required knowledge is reached, but that’s no longer explain something in simple terms, it’s just delaying the proper explanation to when the audience is prepared to tackle it.
i have had plenty of people confidently explain things in a simple manner to me, but they have no idea about what they are talking about. humans and other entities do it online all day long.
It is possible (although unlikely) that someone could explain something to you without understanding it themselves. For the most trivial example, you can imagine a television host reading something off of a teleprompter. I suppose you could counter that such things do not count as “explaining” though.
That's the converse of the claim, which is (understanding something) => (being able to explain it in simple terms). This doesn't imply (being able to explain it in simple terms) => (understanding something) which is what you're considering.
And yes if some topic needs tensor or whatever abstract algebra structure to get it, there is no way to escape learning it first, though the way to teach it might be from tailored gentle companionship to a harsh read the dry book on the matter disconnected from any contextual presentation.
In french, we have this sentence that sums that idea pretty well :
"Ce que l'on conçoit bien s'énonce clairement, et les mots pour le dire arrivent aisément"
Nicolas Boileau (author)
It's worth noting that simple terms needs to be relative to make sense. And I always though this quotation was a proxy for "explaining something minimally and naturally". Because I can explain prime ideals very well and I understand them very well but it wouldn't be considered simple for the average person.
Personally, I think this quotation, being a proxy for a different phenomenon, is inaccurate and asinine.
This is a simple versus easy problem. In the fewest words:
* Simple - numerically fewer
* Easy - least immediate effort
Don't over think that. Simple is not easy. There are many people who cannot express the effort to arrive at simple. I have a kiddo with ADHD/ASD and frequently he cannot arrive at simple. Sometimes the path to easy is a pathological lie that results in greatly extended effort. That inability to arrive at simple could be due to the perceived extended effort required or due to a disgust of executive functioning. It matters almost nothing what the motive behind that inability is as the output is the same.
One of the most important things I have learned from a career in software is that narcissists almost universally cannot measure abstract things with numbers. The required effort is too expensive and too differential to their interests. If you want to know if somebody is a sociopath give them a task with two outputs: the first output requires very little effort and looks amazing, but the second output requires a bit of extended effort and does not look impressive aside from completely dismissing the other output. I have seen this come up dozens of times for numerous people.
What's most interesting about the above paragraph is that narcissists cannot introspect, so they cannot see the obviousness of the above test even when repeated frequently. Everybody else sees this clearly even when they don't realize the absurdity of what they are looking at.
First, you can know topic intrinsically but just be bad at communication. The knowledge of the topic and communication skill to present it are two completely unrelated skills. The people that originally peddled it just couldn't imagine not being great at communication.
Second no, you can't explain complex term to someone that misses every basic piece of knowledge that is required to understand even the basics. That doesn't mean you don't understand it, that just means the topic is very complex.
You can simplify it (to the point of uselessness usually) so the other person have inlking what are implications but not, but they won't exactly know it.
> you can't explain complex term to someone that misses every basic piece of knowledge that is required to understand even the basics
Yes you can, you just need to explain every basic piece of knowledge to them first. Because if you truly understand it, that means you understand all those pieces of knowledge (to a certain degree, of course you'll need to look stuff up, but you'll know what you need to look up, etc.).
That doesn't mean you'll actually have to explain it to somebody with no background knowledge in practice (it would take a really long time), but you yourself was once a person with no background knowledge.
As Feynman says
> “I couldn’t do it. I couldn’t reduce it to the freshman level. That means we don’t really understand it.”
Note that he here targets the freshman level, not somebody with no schooling at all. But of course he could explain it to such a person, he'd just have to bring them to the freshman level first.
Communicating effectively although important - is a different skill than understanding that thing. Understanding is usually communicated by exhibiting that skill in real world. Having a great understanding of something and being able to translate that understanding into words are different things. For example, plenty of senior engineers often struggle with this.
Coincidentally this take usually comes from people who are great at communication but often have very little understanding (often seen in low-skill executives who rise due to brown-nosing / sheer people skills for the lack of a better word)
Being good at communication only means you are good at communication. Nothing more.
plenty of very skilled & smart people are bad teachers of the very thing they excel at.
when concepts are handled at high speed on the daily they become part of your neural circuitry independently of speech or even words, doesn't mean you suck at it, it just means you never had to explain it.
It is very common to hear people express things in jargon these days. Words such as "stakeholder", "resilience", semi-abusive generational labels and we also see the rebranding of words such as "actor" & "narrative" being used to blur the line between truth and fiction.
Trying to get someone to explain a concept in other terms can help them get out of cultic thinking and there is a lot of it now. But there is a tendency nowadays in medicine, science and soft sciences to alter terminology on a regular basis. I have a long term condition which to my knowledge has been relabelled at least five times since my diagnosis. I still have the same thing but not necessarily the in-group terminology.
This was very satisfying when taking math at the undergraduate level, because everything was broken down into proofs from first principles, or theorems from earlier courses, so if you just paid attention you could discover how everything's built from the ground up.
Unfortunately the deeper into graduate studies I got the more it seemed that math "lost its way" and less focus was put on taking the time to really understand things such that you yourself was convinced of it working. Not just the proofs themselves but also the definitions, which often seemed to appear "out of thin air". Whoever pioneered the field probably spent years getting the definitions just right, but as a student we got served the final product with very little explanation of the technical details.
In that sense it felt a lot like primary / secondary school math where the teacher would often reply "it just works that way", which to me is a little disappointing.
In undergrad and early grad, you're praised for your ability to be a persnickety auditor of every little rigorous step. It took me too long to realize that later in grad, you're meant to gloss over the details of things up until it's actually necessary (e.g. you identify the paper you're reading is relevant to your research goals, so you need to dissect it carefully).
Producing didactic material is hard, let alone for the bleeding edge of a niche field. Most of the practical mental models and convincing intuitions you describe happen at conferences and the like, precisely because they're personal, imperfect, and don't really fit in the publishing world. I too wish that knowledge could be collected and disseminated more accessibly.
> you're meant to gloss over the details of things up until it's actually necessary
Right, that's of course true. My feeling was rather that many people will skip this step:
> you identify the paper you're reading is relevant to your research goals, so you need to dissect it carefully
Or at least have a different understanding than me of what "dissect carefully" entails.
> Most of the practical mental models and convincing intuitions you describe happen at conferences and the like, precisely because they're personal, imperfect, and don't really fit in the publishing world.
I'd argue that the point of publishing math is to explain the results to the world, not just make some cryptic note of "I proved this thing, trust me bro (or come talk to me or derive the proof yourself)". But, it seems that's a view held by very few people who are actually working with math, so I suppose I might just not understand them properly. In any case, I'd say that including even an imperfect intuition of what the author is going for, is better than publishing as succinct a paper as humanly possible.
I think this idea is often misunderstood and used as an excuse by people who don’t want to make any effort to understand a complex concept and want to blame the results of their laziness on the explanation or the explainer.
There are things where you may be given an explanation in simple terms, but the concept requires effort on the part of the recipient of the explanation to really understand. I’ll give an example: compactness. Probably anyone who has done analysis is familiar with how this concept is usually taught. Typically how it goes is you do open and closed sets and then they hit you with “a set is compact if and only if every open cover has a finite subcover”. Now this definition is extremely profound but very dense and it’s highly non-obvious why that in particular is the property you want (a finite subcover for every open cover)[1], how you could go about proving that every possible open cover in a certain situation could have a finite subcover etc etc.
In this case it’s actually quite easy to give simple explanations of the separate parts of the definition (open and closed sets, what an open cover therefore means, what a finite subcover then is etc) but actually understanding the concept of compactness requires something above and beyond whatever explanation you are given: it requires the reader to get out pencil and paper, try out some sets, go for some walks, think about it in the bath etc.
You may get a really fantastic explanation in simple terms from someone who deeply understands the topic, but if you don’t put in the effort you’re not going to understand it yourself. That’s not the fault of the explanation. It’s not an easy concept to grasp so you have to climb the ladder of understanding. The explainer can’t do that for you.
[1] it took the mathematical community years to settle on this being the definition, so you shouldn’t expect to understand why this is the definition straight away.
The article is just a bunch of quotes from Richard Feynman and John Gruber. There is little insight and I even think the article is misrepresenting the quotes.
Feynman said "if you can't explain it on a freshmen level". Leaving out that qualifier distorts the statement.
The other quote about the magnets is also overly reductive. Saying "I don't understand it in terms you're more familiar with" is not an admission that scientists as a group don't understand it at all.
You could go into what it means to understand something, but unfortunately the article isn't doing that.
Besides, the claim on magnetic force already serves as a counter-example to the main claim of the article.
Unless the article means to say that Feynman didn't understand the magnetic force, they have to admit that the magnetic force was perfectly understood, but impossible to explain in simple terms.
Explain to what level? E.g some topics common here. How do you explain NP in simple terms? "Things that are hard to compute?" That's simple, but not much of an explanation. It leaves the listener with a practically useless definition. And how do you explain even something practical as Heartbleed or Rowhammer? Etc.
And then there's another barrier, of course: the audience. You may use simple terms, and tell a lengthy story, but some things are hard to understand. Relativity and Monty Hall come to mind. They can be explained in simple terms, but how many people get them? Probably just the ones that also understand a detailed explanation.
Simple for your audience to understand. Which means you posses understanding deep enough that you can explain it any any level of detail.
So, should you also understand your audience, too? Is that ever possible? What if there's 300 people who listen to you with their own worldviews, viewpoints, health conditions (see different colors), life experience?
Literally yes, as much as possible. When preparing for an audience, you generally want to consider who your audience is when you're preparing your communication.
It doesn't matter what you know or understand, if you show up to a Spanish class and give a lecture in French.
I think the formulation staying at an audience of 1 gets the point across.
“Also you have to be good at educating groups” doesn’t add anything meaningful IMO
I don't think explaining NP is that bad. Most people can understand what an algorithm is, some basic problems like sorting a list or TSP. If you just want to distinguish polynomial-time from non-polynomial-time algorithms, it's straightforward to give "obviously bad algorithms" (like brute-force search) and "obviously better algorithms" (pick your favourite) and to make that intuitive distinction precise.
Practical things like computer vulnerabilities are pretty easy to explain. I think anyone I know could understand what a buffer overflow is and why it's bad. The computer is a dumb machine and clobbers whatever you tell it to.
Relativity and monty hall are much better examples, I think. Relativity is well-studied enough that the explanations available for teenagers are pretty good, and if you study them you can get a pretty comprehensive understanding. But you do have to study them. Both are vulnerable to a certain kind of passive understanding, where correct arguments and incorrect arguments can both sound equally convincing.
E.g. NP is a problem that's easy to check, but not easy to solve. NP is type of a problem like one for finding people: it's easy to confirm (is Mr. X in the room?), but there's no quick way to find solution (where is Mr. X).
All of these remarks are addressed in the article.
That means you don't really understand it
Lets put it this way - you shouldnt be a teacher probably if you can't even wrap your head around this while everybody else seemingly can.
I have yet to meet a single problem that when broken down to atomic parts wasn't trivial to grok. Every single piece of software. I am not saying 3 year old should understand details of general relativity, thats not the point.
Erm... I think you've gone too far.
> I have yet to meet a single problem that when broken down to atomic parts wasn't trivial to grok.
For one thing, we don't know how to break every problem down to "atomic parts". More seriously, the important things to grok in life are the things that are not atomic parts! If you look at quantum mechanics as a system of fancy mathematics with arbitrary rules, it is possible to become an unthinking machine that calculates outcomes. But that is not at all what it means to "grok" quantum mechanics! Likewise with relativity - it is very easy to have Lorenz transformations memorised but be totally unable to explain how e.g. the twin paradox works. Saying that's trivial to grok is utter BS.
The problem with your NP explanation is not the level, but that it is wrong, based on a common misunderstanding. The correct simple explanation is:
NP = "Things that are easy to verify"
What you were thinking was probably the concepts of NP-hard or NP-complete.
Who the fuck is explaining NP to their team on a daily basis? Come one, stop being a pendantic little bitchb
It seems like you don't understand those things. I can envision explaining them in simple terms. NP isn't very complicated. xkcd had a great Heartbleed explanation in visual form https://xkcd.com/1354/
4/5
Spent a while looking at that. Not a clue
I don’t think this is true because it makes an assumption that reductionism is always possible, you can always distill every increasingly complexity into something less and less complex and maintain the overall behavior. I don’t think complexity is always reducible because sometimes the detailed nuance is absolutely critical to the distinction.
You may be able to take core ideas from something and distill them but when you have multiple interacting core ideas you may not necessarily be able to reduce them to simple terms. And almost always when you do there’s a big asterisk that the analogy has a lot of flaws.
This is exactly what the article says. If you find it hard to explain in simple terms then you do not understand it enough.
Doesn’t mean it’s true
So Feynman didn't understand magnets enough?
It may be that sone things can’t be explained in simple terms, but perhaps they are just not understandable?
I think it's not really about making simple analogies, because they don't really "explain" anything in the proper sense of the word. An analogy might make somebody feel like they understand it, but it doesn't really give actual understanding (see pop sci explanations of quantum mechanics).
It's more about breaking down stuff into simpler parts. For example, you want to explain a buffer overflow exploit in a web server. Then you'll ask your audience, do you understand addressing / memory allocation, and do you understand HTTP? If they say no to any of those, then you'll have to explain e.g. memory allocation first. And if they don't know how a CPU / RAM / caches operates, then you'll have to explain that first. And maybe you'll have to explain boolean algebra and how to build an adder from logic gates. And so on.
So I would claim that "reductionism is always possible". More complex things just take longer to distill into less complex things, but it can always be done. The exception would be things that "nobody really understands", like Feynman says about magnetism:
> I really can’t do a good job, any job, of explaining magnetic force in terms of something else you’re more familiar with, because I don’t understand it in terms of anything else you’re more familiar with.
Well I would push back on that claim because I don’t feel like you’re reducing things. You’re filling in gaps in understanding of something they already understand across the entire conceptual landscape it’s just masked by prior knowledge, they already know other concepts in detail and using a strategy to break a larger problem into smaller parts to make it digestible over time, you’re still dealing with the full set of complexity.
I think reductionism absolutely is possible in a lot of cases but I tend to agree with Stephen Wolfram’s concept of “computational irreducibility” or concepts we essentially know from compression in information theory (like Shannon entropy, Kolmogorov complexity) that there are limits to just how much you can reduce an idea or concept, and the bound for the human mind is likely much higher than those lower limit concepts.
Magnetic force is pointed out because it’s a more unique experience people don’t have a lot of prior knowledge and experience baked in that you can leverage as an assumption to reduce the concept, I think this holds true with lots of things. That isn’t to say we shouldn’t try to explain things simpler, just that that big caveat asterisk becomes ever more important when we use reductionist explanations. It’s not unlike when people try to explain spacetime curvature with a ball on a rubber sheet and things rotating around it… sure, that kind of gets at the idea, but there’s a huge asterisk that it’s simply not what’s really going on, it hints at the concept of curvature on a 2D plane you need to somehow conceptually extrapolate to other dimensions.
CS Lewis, an academic in a completely different field (literature: Professor of Medieval and Renaissance Literature at Oxford), said something very similar.
Being able to break things down into simple explanations (maybe many of them though?) is a sign of rigorous thinking. Not being able to do so (which seems common in some fields) is a sign of the opposite.
"How does weight loss work?"
Calorie Deficit
"Does it work on all people?"
Yes, without a doubt.
"But humans are not bomb calorimeters."
Doesn't matter; minor differences in metabolism/absorption <<< calorie deficit.
Many people know X matters, Y matters, and Z matters, but very few are able to quantify how much X matters in relation to Y or Z.
This is why many people miss the forest for the trees.
Everyone working on how Claude communicates would benefit from reading this.
Yes, this is often obscured when working with code because it's relatively easy to judge its correctness. When using Claude to write documents, describe a system or organize requirements it becomes clear very quickly that Claude doesn't really understand what it is writing, it just puts probable things close together.
Doesn't mean Claude isn't super useful though.
I think the problem here is the coding agent that we are working with are not familiar with the degree of knowledge we have around the system
It's like throw a lot of concepts at once that were just created 5mins earlier but the agent doesn't care that the human supervisor isn't aware about them and how they function, so it doesn't know what simple explanation in relative terms should be provided next
I think this is also implied in the article in the ending paragraphs
There are skills which try to force Claude Code to use Feynman technique like [1] and many others
[1] https://github.com/TheJavaGuy/tjg-claude-marketplace/tree/ma...
Well surely claude struggles here because it doesn't '''understand''' anything.
When making such strong claims that could easily devolve into semantics, it would be valuable to explain what you understand to be the definition of the semantic fulcrum of the claim.
In this case, that semantic fulcrum is what it means to '''understand''' something. Because the question of whether Claude understands things hinges much more on the criteria for understanding, than it does on our belief on the capabilities that Claude has.
i think a big part of my non-progrmamming IT work is pedagogy: many people build systems (even before AI) without being able to explain why/how they work
As everything, I think it depends. When the topic under discussion is complex and requires extensive background knowledge, it’s still possible to explain it in simple terms but the risk is to oversimplify it and loose important details. The alternative would be to analyse every sub-topic until the required knowledge is reached, but that’s no longer explain something in simple terms, it’s just delaying the proper explanation to when the audience is prepared to tackle it.
i have had plenty of people confidently explain things in a simple manner to me, but they have no idea about what they are talking about. humans and other entities do it online all day long.
You have the correlation wrong.
"Everyone who understands X can explain it in simple terms" =/= "Anything explained in simple terms is done so by someone who understands it"
Nice - clearly explained in simple terms.
It is possible (although unlikely) that someone could explain something to you without understanding it themselves. For the most trivial example, you can imagine a television host reading something off of a teleprompter. I suppose you could counter that such things do not count as “explaining” though.
> The internet is a series of tubes [...] and not a big truck.
If they have no idea what they're talking about, then they haven't explained anything to you.
Also, "explaining X" doesn’t mean "explaining Y, where Y is a poorly understood version of X".
That's the converse of the claim, which is (understanding something) => (being able to explain it in simple terms). This doesn't imply (being able to explain it in simple terms) => (understanding something) which is what you're considering.
I think the caveat is, simple to me might not be simple to you.
You need to understand the subject to explain, and the audience you're explaining it to.
https://m.youtube.com/watch?v=SxdOUGdseq4&t=346s&pp=2AHaApAC... comes to mind
And yes if some topic needs tensor or whatever abstract algebra structure to get it, there is no way to escape learning it first, though the way to teach it might be from tailored gentle companionship to a harsh read the dry book on the matter disconnected from any contextual presentation.
still wrong
some thinking and understanding is conceptual, spatial, bound to mind-muscle, smells, etc.
a good chef understands the ingredients' cooking dynamics perfectly and may still have trouble putting this into words
In french, we have this sentence that sums that idea pretty well : "Ce que l'on conçoit bien s'énonce clairement, et les mots pour le dire arrivent aisément" Nicolas Boileau (author)
Maybe words come easily, but maybe they are not simple words, but are words that for instance rely on not well-known concept / abstraction / ...
It's worth noting that simple terms needs to be relative to make sense. And I always though this quotation was a proxy for "explaining something minimally and naturally". Because I can explain prime ideals very well and I understand them very well but it wouldn't be considered simple for the average person.
Personally, I think this quotation, being a proxy for a different phenomenon, is inaccurate and asinine.
This is a simple versus easy problem. In the fewest words:
* Simple - numerically fewer
* Easy - least immediate effort
Don't over think that. Simple is not easy. There are many people who cannot express the effort to arrive at simple. I have a kiddo with ADHD/ASD and frequently he cannot arrive at simple. Sometimes the path to easy is a pathological lie that results in greatly extended effort. That inability to arrive at simple could be due to the perceived extended effort required or due to a disgust of executive functioning. It matters almost nothing what the motive behind that inability is as the output is the same.
One of the most important things I have learned from a career in software is that narcissists almost universally cannot measure abstract things with numbers. The required effort is too expensive and too differential to their interests. If you want to know if somebody is a sociopath give them a task with two outputs: the first output requires very little effort and looks amazing, but the second output requires a bit of extended effort and does not look impressive aside from completely dismissing the other output. I have seen this come up dozens of times for numerous people.
What's most interesting about the above paragraph is that narcissists cannot introspect, so they cannot see the obviousness of the above test even when repeated frequently. Everybody else sees this clearly even when they don't realize the absurdity of what they are looking at.
That would mean that no-one understands quantum mechanics.
Which is indeed an often made claim, including by Feynman.
And its what the Quantum Mechanics adage of "Shut up and calculate" is meant to confer to students as well.
> I couldn’t reduce it to the freshman level. That means we don’t really understand it.
Pretty big ego
Well if this applies to making softwares, then we will have a lot to explain, are we still shipping it or not?
I hate that stupid myth.
First, you can know topic intrinsically but just be bad at communication. The knowledge of the topic and communication skill to present it are two completely unrelated skills. The people that originally peddled it just couldn't imagine not being great at communication.
Second no, you can't explain complex term to someone that misses every basic piece of knowledge that is required to understand even the basics. That doesn't mean you don't understand it, that just means the topic is very complex.
You can simplify it (to the point of uselessness usually) so the other person have inlking what are implications but not, but they won't exactly know it.
> you can't explain complex term to someone that misses every basic piece of knowledge that is required to understand even the basics
Yes you can, you just need to explain every basic piece of knowledge to them first. Because if you truly understand it, that means you understand all those pieces of knowledge (to a certain degree, of course you'll need to look stuff up, but you'll know what you need to look up, etc.).
That doesn't mean you'll actually have to explain it to somebody with no background knowledge in practice (it would take a really long time), but you yourself was once a person with no background knowledge.
As Feynman says
> “I couldn’t do it. I couldn’t reduce it to the freshman level. That means we don’t really understand it.”
Note that he here targets the freshman level, not somebody with no schooling at all. But of course he could explain it to such a person, he'd just have to bring them to the freshman level first.
Communicating effectively although important - is a different skill than understanding that thing. Understanding is usually communicated by exhibiting that skill in real world. Having a great understanding of something and being able to translate that understanding into words are different things. For example, plenty of senior engineers often struggle with this.
Coincidentally this take usually comes from people who are great at communication but often have very little understanding (often seen in low-skill executives who rise due to brown-nosing / sheer people skills for the lack of a better word)
Being good at communication only means you are good at communication. Nothing more.
Couldn't have made a more lazier shallow take.
More lazier
ITT: people not realising they don't understand things.
5/5
I'm genuinely surprised at the amount of pushback this idea is getting here.
because it's wrong
plenty of very skilled & smart people are bad teachers of the very thing they excel at.
when concepts are handled at high speed on the daily they become part of your neural circuitry independently of speech or even words, doesn't mean you suck at it, it just means you never had to explain it.
ITT: people confusing understanding something and being able to convey the fact you're understanding it to others
yet to hear any explanation of quantum computing, that makes any sense....
1Know it exists.
2Know how it’s used.
3Understand it.
In that order.
You don’t need to understand everything. Delegation is the greatest invention of humanity.
Explain to me how to become billionaire
Be born with it, marry into it or try work your entire life for it and succeed 0.0001% of the time.
It is very common to hear people express things in jargon these days. Words such as "stakeholder", "resilience", semi-abusive generational labels and we also see the rebranding of words such as "actor" & "narrative" being used to blur the line between truth and fiction.
Trying to get someone to explain a concept in other terms can help them get out of cultic thinking and there is a lot of it now. But there is a tendency nowadays in medicine, science and soft sciences to alter terminology on a regular basis. I have a long term condition which to my knowledge has been relabelled at least five times since my diagnosis. I still have the same thing but not necessarily the in-group terminology.
This was very satisfying when taking math at the undergraduate level, because everything was broken down into proofs from first principles, or theorems from earlier courses, so if you just paid attention you could discover how everything's built from the ground up.
Unfortunately the deeper into graduate studies I got the more it seemed that math "lost its way" and less focus was put on taking the time to really understand things such that you yourself was convinced of it working. Not just the proofs themselves but also the definitions, which often seemed to appear "out of thin air". Whoever pioneered the field probably spent years getting the definitions just right, but as a student we got served the final product with very little explanation of the technical details.
In that sense it felt a lot like primary / secondary school math where the teacher would often reply "it just works that way", which to me is a little disappointing.
In undergrad and early grad, you're praised for your ability to be a persnickety auditor of every little rigorous step. It took me too long to realize that later in grad, you're meant to gloss over the details of things up until it's actually necessary (e.g. you identify the paper you're reading is relevant to your research goals, so you need to dissect it carefully).
Producing didactic material is hard, let alone for the bleeding edge of a niche field. Most of the practical mental models and convincing intuitions you describe happen at conferences and the like, precisely because they're personal, imperfect, and don't really fit in the publishing world. I too wish that knowledge could be collected and disseminated more accessibly.
> you're meant to gloss over the details of things up until it's actually necessary
Right, that's of course true. My feeling was rather that many people will skip this step:
> you identify the paper you're reading is relevant to your research goals, so you need to dissect it carefully
Or at least have a different understanding than me of what "dissect carefully" entails.
> Most of the practical mental models and convincing intuitions you describe happen at conferences and the like, precisely because they're personal, imperfect, and don't really fit in the publishing world.
I'd argue that the point of publishing math is to explain the results to the world, not just make some cryptic note of "I proved this thing, trust me bro (or come talk to me or derive the proof yourself)". But, it seems that's a view held by very few people who are actually working with math, so I suppose I might just not understand them properly. In any case, I'd say that including even an imperfect intuition of what the author is going for, is better than publishing as succinct a paper as humanly possible.
80IQ idea.