Sal Khan: The amazing AI super tutor for students and teachers [video]
ted.com
ted.com
Whom will the AI teacher teach then?
Even if my own job gets replaced, and I can no longer find employment, I will keep learning new Math, new programming languages, and other things. (Seems unlikely now, but definitely not impossible)
But I have seen that people like me are rare.
Most students learn with an earthly, material goal in mind. And you can't blame them.
Life in the Upper West Side, NY, will not be the same as a Bangladesh lower middle class neghborhood.
A lot of jobs providing people with middle class incomes will be wiped out. The middle class will drastically shrink. That is the scenario that is the most likely.
Meanwhile the rich will no longer have any need for the poor, and since 100% of your worth as a human being derives from what job you do in service of the rich, you'll become a worthless little fucking bug for them to exterminate.
Don't get me wrong: I like your vision better.
This matches the data.
First of all, prove that. Second of all, prove they won’t be replaced by more interesting, more engaging jobs.
Then you might actually have a point instead of just FUD.
Letting the machine babysit our kids so we can... what? Do work that a computer could do? Party all day? Why are teachers so poorly paid?
https://en.wikipedia.org/wiki/Bloom%27s_2_sigma_problem
Individual tutoring plus mastery (which Khan already has) outperforms classrooms by 2 sigma. Coming up with ways that we can individualize education which we can afford (since we're not going to buy individual tutorial education for every child) would be huge.
Next year, I'm helping a group of ELA teachers use GPT4 to provide each student with individualized writing feedback. We have a relatively low student to teacher ratio, but still we can only provide this kind of depth 3-4 times per year, and now we're going to get it a couple times more.
And in the microeconomics class I taught, I enjoyed the efficiency and educational aspects of having about a quarter of my assignments be on Khan. Exercises are self-grading; students get immediate feedback; and if students need extra help, it's there... with Sal Khan explaining things radically different from how I did.
(I like my explanations better, but anyone who didn't click with things my way had a chance to try something different).
...by what metrics? Anyway, who cares? Why do kids need an education, let alone one that has sigmas, in a world where talking machines can outperform them in every area?
What are people for? (That's a Wendell Berry essay, also I'm asking.)
I maintain that the robots should do the work so that people can e.g. teach their own kids. Education should be more than a means to secure a position in the economic/social order. (An order that seems poised for some sort of upheaval.)
The Wikipedia article is linked right there.
> in a world where talking machines can outperform them in every area?
Seems unlikely for awhile, except perhaps as a self-fulfilling prophecy.
> I maintain that the robots should do the work
It sure would be nice if robots could like, unclog toilets and clean, instead of usurping creative work.
> so that people can e.g. teach their own kids
Some people are really well equipped for this, in knowledge, temperament, and ability. Others, ehh..
Yeah, thanks. It says "a knowledge test"...?
Anyway, this Bloom is clearly a crank:
> Mastery learning is an educational philosophy first proposed by Bloom in 1968[8] based on the premise that students must achieve a level of mastery (e.g., 90% on a knowledge test) in prerequisite knowledge before moving forward to learn subsequent information on a topic.
That's not how brains work. No wonder the educational system is so pathetic, eh? I recall when "points" were introduced in school, and even as a kid I could recognize that this was a serious degradation of educational philosophy and technique.
>> in a world where talking machines can outperform them in every area?
> Seems unlikely for awhile
It's already happening. Jobs are evaporating. Opportunities for gainful employment are withering under the glare of the LLMs as we speak. I just checked Craigslist and there are no jobs for me there anymore.
> It sure would be nice if robots could like, unclog toilets and clean, instead of usurping creative work.
Yeah, that's my whole point here: "Let the robots do the work and we'll take their pay."
People talking about robot judges and robot teachers, etc., are putting the cart before the horse. I remember when Sony made that dancing robot and I was like, "Damn it, I like to dance! Make a robot that does the #$@% dishes!"
>> so that people can e.g. teach their own kids
> Some people are really well equipped for this, in knowledge, temperament, and ability. Others, ehh..
"It takes a village...", eh? The "unit" of human is not the individual, or even the nuclear family or extended family. It's the village/neighborhood.
Teaching kids to be adults is not exactly the hardest thing to do? We managed it for about a million years before we ever even lived in towns.
It's been reproduced on large numbers of measures of performance-- e.g. on performance on a mathematics test at the end.
> That's not how brains work.
There's a lot of evidence for mastery-based instruction-- which attempts to keep students working on topics they have not quite mastered-- instead of methods that advance students to material that depends upon earlier material that they've not demonstrated mastery of. It seems perhaps you're quibbling with the threshold-aspect as implemented?
It's kind of annoying when people dismiss an idea without taking time to fully engage with and understand it. I expect more from my middle school students, let alone someone I'm taking time to discuss things with on HN.
There's a pretty big problem in classroom style education, where students fall behind once and become accustomed to being confused. It's very, very difficult to claw back onto the line of high performance once this happens to you. There's another problem, too, in that even if you have tiered classes there's quite a lot of variation within a class and quite a lot of variation even with the high performing students on how quickly they grasp each idea. Mastery methods attempt to modulate pacing for each student for optimal outcomes.
The difficult thing with mastery based methods in a normal classroom is it's difficult to keep giving meaningful lessons and coaching when you have students before you at 5 different levels of mastery and thus different places in the curriculum. It's still better than blindly advancing everyone based on where the 40th percentile of the classroom overall is performing. But the hope is LLMs can pitch in and help here and provide a little bit of differentiation and individualization.
> We managed it for about a million years before we ever even lived in towns.
We have much higher expectations now than we did a million years ago. We weren't trying to get a relatively big fraction of the populace to understand calculus, write eloquently, and reason deeply in the social sciences.
I fully agree, and I wish you'd do your part. What are people for?
It's a quest pretty much everyone struggles with. And it's one that's made harder by everything getting so big and mechanization taking away so much. But there's no panacea in returning to an agrarian, village life as you seem to suggest.
Many of us enjoy the vibrancy of the larger world, and enjoy the bigger things we are able to accomplish by educating ourselves (and our children) deeply, and also enjoy our deeper engagement in this larger complicated world. Why should we not listen to educational research (other than we have precious little of it to listen to)?
I know I am easily fooled by what "seems to work" and the power of the anecdote. I seek to continually improve through careful study of what can be proven.
We know that we don't have the resources as a society to provide the depth of education we currently do in a one-to-one or co-op format. We also know the quality of real co-op or homeschool is terrifically variable. Some of the best education is there. Much more of the worst is.
(If you haven't, do read Wendell Berry's "What are People For?" or any of his work. His writing is so fine!)
> there's no panacea in returning to an agrarian, village life as you seem to suggest.
I agree, I'm pointing out that complexity itself has become the problem of the day (in my opinion.) FWIW, I just acquired some land on which to experiment with a high-tech, ecologically-harmonious lifestyle. Fingers crossed. When I was a young man I read in an old Taoist book about a farmer who was so content with his life that he had never even been to the next village over, even though he lived so close to them he could hear the cries of their rooster. At the time it gave me a feeling of dull horror. Now, nearly fifty, I envy that farmer and hope to reach his level someday before I die. Times change.
> Why should we not listen to educational research (other than we have precious little of it to listen to)?
Well, there it is: paucity of reliable data and research. Further, and to my point (why I'm jerking your chain) there's the problem with the very idea that education can be researched.
The growth of the human mind cannot be represented by a single scalar value. At least not for any admirable end, eh? It doesn't matter whether that scalar value is graduated in five increments ("A" through "F") or fifteen (A+, A, A-, ... F-, does anyone actually use F-? Hehe.) or a hundred (per cent), it's just silly.
It was understood, at least I formed the impression as a child, that the A-F grade system was a dufus solution to a problem nobody was actually trying to solve. Schools are part of the anthropological machinery that societies use to get the kids to behave like the adults. Any educational value they have is a side-effect.
You cannot actually educate a child outside of a deep, sustained (20-30 years), one-to-one relationship between a caring adult who is sane mentally and emotionally and the child. [edit to add: well, really I suspect it takes at least a whole village, eh? Where are you gonna find a whole village of mentally and emotionally sane people on Earth in this day and age, huh?] To a first approximation this is pretty much the most important human activity. All others rely on it, eh? Education is the process of growing humans out of talking primates.
So again, the mind that conceives of measuring the "mastery" of the human mind using a single scalar value is the mind of a fool, and in my view is automatically disqualified from having their views and ideas considered at all, let alone taken seriously. He's not "sane mentally and emotionally" and should not be let anywhere near children or schools!
To me, once you get away from physics, most science seems to be a kind of cargo-cult. Chemists and biologists are mostly okay, but sociologists, psychologists, economists, etc. are not actual scientists in my view. They do not do science, they ape it. So maybe you want to back away from this conversation? I won't be offended. That's why I told you. So you don't waste your time. :) (I also "believe in" Neurolinguistic Programming.)
But getting back to my point: you can't measure learning by a single scalar value; the idea is pernicious and counter-productive.
Corollary: people who believe such a silly thing, and who somehow get a hold of kids to experiment on, will go right ahead and plug them into talking "learning machines", and they will believe they do it "for their own good". Not to put too fine a point on it: your life is my horror story.
(I was educated in public school in San Francisco. Fair warning: you're not going to be able to change my mind about how much of a bad idea modern schooling is. I'm still feeling the effects three decades later.)
I think you've fundamentally misunderstood the concept. Mastery levels are a huge vector of how well you've demonstrated various skills. The main idea is that you don't move on to a later skill until you've shown high proficiency in the skills that it depends upon.
Of course, there's always room for judgment and to adjust things based on individual students.
> You cannot actually educate a child outside of a deep, sustained (20-30 years), one-to-one relationship between a caring adult who is sane mentally and emotionally and the child.
In the end, educators do depend upon the parental relationship to be successful. But I get an awful lot done with excited students in several weeks of 40 minutes per day (average).
> (I was educated in public school in San Francisco. Fair warning: you're not going to be able to change my mind about how much of a bad idea modern schooling is. I'm still feeling the effects three decades later.)
I didn't enjoy my school years much. That's part of why I'm here and teaching. Plugging my ears and going "la la la" about the evidence and just doing what I want is not how I will make anything better.
> there's the problem with the very idea that education can be researched.
You can definitely divide students into groups, and do different things, and measure outcomes. We have that for mastery, for tutorial methods, for some things like "flipped classrooms," etc.
You can also do case control studies, though you need to be very cautious because of confounds.
It's worse than that: I've not even tried to understand it.
> The main idea is that you don't move on to a later skill until you've shown high proficiency in the skills that it depends upon.
Okay. That idea is foolish and regressive. The human mind learns on multiple levels simultaneously. A single "lesson" can evoke or participate in many "learnings" over the course of a pupil's life. You can teach a skill and the skills that depend on it at the same time, if you know what you're doing.
Arguing the details of how Bloom is a crank is uninteresting to me. I would sooner study the "evidence" of the free energy generator crackpots, as I would be slightly less certain that I wasn't wasting my time.
You don't know what you're doing. Doing it harder and faster with computers is not a good idea. I pity your student and you as well.
> Plugging my ears and going "la la la" about the evidence and just doing what I want is not how I will make anything better.
If you want to discuss how the weird shit you're doing to these kids will or will not prepare them for the world where computers talk and think better than they can, I'd love to do that. If you want to passively condescend to me because I won't taste your particular flavor of pseudo-scientific kool-aid I've got better things to do with my time.
Thanks for admitting bad faith before going straight into ad-hom!
You're clearly intelligent and sincere, which is why I bothered trying to reach you, so it's disappointing that this turned out to be a waste of time, even though I feel it was worth the attempt. As I said, I believe education is pretty much the most important human activity, so even though I have no children of my own. I feel it behooves me to do what little I can for the children of the rest of humanity.
I am discussing this in good faith, and I am not trying to insult you. However, I do sincerely believe that you are deeply confused about the nature of education, and I'd very much like to help you clear that up, especially since you teach kids. But I can't help you until you give up this crazy idea that "research" has provided "evidence" that your Mister Bloom's weird ideas about education (et. al.) are, uh, not poppycock.
(I mean, I did say, right off "...by what metrics? Anyway, who cares?" I feel I've been very clear this whole time that I'm not interested in Bloom's work because to me he's obviously a crank. You keep trying to defend him when I'm not even willing to engage. Your "evidence" means nothing to me because I don't believe it's actually scientific. IF you wanted to convince me to take it seriously you would first have to convince me that it has any meaning at all. And you're not going to be able to do that until you can show me that you understand the nature of my objection, which you haven't done yet.)
I've not been talking about Bloom except for pointing out his initial observation about the difference in outcomes with different instructional methods.
The rest of the time has been defending simple ideas like "we can empirically study education."
If you're taking me explaining the concept of mastery as a defense of Bloom, then we've really gone awry. It's not really a Bloom idea (though I guess he had a little influence on modern understanding of the idea).
But you keep bringing him up in every other comment and ascribe my comments as a defense of him. The really wack thing is, Bloom's observations would really support your point (he measured that tutorial instruction was more effective than classroom instruction which is a portion of the point you're seemingly trying to make).
You clearly have really big feelings about education-- that I think are influencing you and preventing you from engaging meaningfully with anything I say. Either that, or lack of reading comprehension has gotten in your way. In any case, it's hard for me to see the point in further conversation.
It's too bad, too. The classroom that I teach in is not at all like the classrooms I grew up in and did not enjoy. Perhaps I have some meaning I could share, but you dismissed me at message 1 and haven't tried to understand me at all; instead you're painting me with the same brush as your negative experiences of the past.
The average SAT score across all people entering teaching colleges is ~1060.
When considering whether public school teachers are underpaid on average, and comparing their pay to that in other roles:
1. consider that they work for only 38 weeks per year (instead of 50)
2. consider that in many states, teachers are entitled to defined-benefit pensions and post-retirement healthcare
3. compare pay to other professions with similar entry standards (i.e. what other professions have average SAT scores of 1060).
#1 is equivalent to a pay increase of ~30%
#2 is equivalent to a pay increase of ~45%
So, take the base salary, multiply it by 1.89 (1.3*1.45), and then compare it to non-government jobs that you can get with similar academic performance.
You're also ignoring credentialing, continuing education, and total education requirements.
In my case, I work 40 weeks a year. I work at a private school and just have a 403(b) plan which has less favorable matching and vesting than most 401(k) plans. My total comp is well under 10% of what I could make in private industry. It's not exactly easier work to do well, either.
In this area (SF Bay Area), it's not really a living wage unless you have a spouse that can subsidize you.
In the end, the low comp is causing a lot of teachers to self-select out of the profession, and it's a reason why we have a teacher shortage. If the pay were as ample as you suggest, this would not be happening.
An equally big reason is the low social status and often-poor working conditions of teachers.
The public spends boatloads on education, but in the public sector, a bigger and bigger administrative class and dubious, administrative-driven spending gobbles it up.
When considering whether public school teachers are underpaid on average
> you're valuing #2 very highly; it is worth 45% to very few educators
Take a look at this page of actual salaries for SFUSD, and actual total pay and benefits for SFUSD teachers: https://transparentcalifornia.com/salaries/search/?q=Eld%20C...
Total pay is 43% to 153% higher than base pay. The average % is 55%.
I'm not cherry picking. I just picked a job role that people think is underpaid (elementary school teacher), and looked at the first page of results.
> In the end, the low comp is causing a lot of teachers to self-select out of the profession, and it's a reason why we have a teacher shortage.
You're probably right about this. But I suspect this causes the most capable people to leave (or just never consider entering) the profession. I'm in favour of both raising the bar for teachers, and paying them more to attract people who can meet that bar. Unfortunately, it seems like districts and unions are unwilling to hold teachers to account[0]. (Yes, I know districts have many other failings that are more severe than this.)
> You're also ignoring credentialing, continuing education > and total education requirements.
Honestly I'm skeptical of the value of some of the stuff required for teachers to become and stay credentialed. And teaching is far from the only job that requires continuing education.
> My total comp is well under 10% of what I could make in private industry.
This is really interesting. Do you mind me asking how you look at this decision? e.g. why not take the job paying 10x what you make now, and use some of the excess money to help students in other ways? Or take a part-time job paying 5x, and volunteer the rest of the time?
[0] https://en.wikipedia.org/wiki/Waiting_for_%22Superman%22
You're showing the top of a table sorted by total comp in a particularly well paid district (as property tax revenues pay teachers). This is not representative.
The top entry in the table was occupying an administrative role.
> I'm in favour of both raising the bar for teachers, and paying them more to attract people who can meet that bar.
You can't raise standards while you have a shortage. You need a surplus to be selective.
> And teaching is far from the only job that requires continuing education.
Yes, but it's rare in the cohort of jobs you are seeking to compare against.
> This is really interesting. Do you mind me asking how you look at this decision? e.g. why not take the job paying 10x what you make now, and use some of the excess money to help students in other ways? Or take a part-time job paying 5x, and volunteer the rest of the time?
I enjoy it and it feels more impactful. This is why teachers accept being underpaid and having worse working conditions.
This is how I chose that link:
1. Go to the web site.
2. Enter the name of a friend who is an SFUSD elementary school teacher.
3. Click on the job title.
Whilst SFUSD may not seem representative to you, it seems representative to me, because:
- it's my local district, and where I had planned to send my son to school
- there are often cries that SFUSD teachers are underpaid, can't afford to live in SF etc.
I'd encourage you to share a different link, that shows evidence that my 45% claim is off.
> You can't raise standards while you have a shortage. You need a surplus to be selective.
Yes you can, if you have to be willing to relax two assumptions:
- public school teacher pay must be based only on tenure and credentials (i.e. you can't pay new teachers more than existing teachers)
- new teachers must be willing to spend a year getting a teaching credential (i.e. no one can easily switch jobs from a different profession)
These two are preventing public schools from tapping sources of great teachers (e.g. folks like you, with industry experience and a love of teaching).
Imagine if Sal Khan were to apply for a teaching position at SFUSD, and was willing to work for $150k base. They wouldn't hire him, because:
- he doesn't have a teaching credential (AFAIK);
- even though he might be as effective as two existing teachers, union contracts determine pay ranges.
> I enjoy it and it feels more impactful.
Yup, 100% makes sense.
You still have a table sorted by comp, where the top many rows include teachers who have taken on administrative duty, worked over summer, etc.
> Whilst SFUSD may not seem representative to you, it seems representative to me, because:
Sorry, this doesn't fly. You made a blanket statement about public school teachers, not public school teachers in SF.
> - there are often cries that SFUSD teachers are underpaid, can't afford to live in SF etc.
And the lower rows in those tables, making $45k base (or less) plus benefits, are going to have a pretty hard time doing so. And, SF is still a much more desirable working environment and comp environment than most public school districts.
> > You can't raise standards while you have a shortage. You need a surplus to be selective.
> Yes you can, if you have to be willing to relax two assumptions:
No, if you are going to pull people into the profession from elsewhere, you're going to need to have higher average comp than now. You probably have to also significantly improve working conditions. After all, us private school teachers are willing to make 30% less for better working conditions.
Another faulty assumption is that academic performance is the key figure of merit for educational effectiveness. If you're teaching computer science or calculus, maybe it's very important. On the other hand, this is probably not true for elementary teachers. And while salaries overall correlate to SAT scores, it is a very poor predictor of salary in most fields.
You questioned my 45% assertion. I provided some data. You say it's not representative. I invited you to provide different data that contradicts my point. Again, I invite you to change my mind by showing me some more representative data.
I checked another district in California (Morgan Hill). Looking at all 2021 employees whose job title includes the word 'teacher', the median additional compensation is 21% of base salary. I'm not familiar with this district, so I don't know whether those numbers include the present value of unfunded pension liabilities. If so, the actual number is higher.
> And the lower rows in those tables, making $45k base (or less) plus benefits, are going to have a pretty hard time doing so.
Those numbers are total compensation in a particular calendar year, and may be for employees who worked part time or joined/left part way through the year. No elementary school teacher in SFUSD has a base salary that low.[0]
> No, if you are going to pull people into the profession from elsewhere, you're going to need to have higher average comp than now.
This doesn't contradict what I said. One of the things I said (paraphrased) is that you have to be willing to pay new teachers more, without being constrained by the salaries of teachers hired under the old system (with the old bar).
> Another faulty assumption is that academic performance is the key figure of merit for educational effectiveness.
I'm not sure where this point is coming from. I never stated or implied this assumption.
I did refer to academic performance (as estimated by SAT score). Academic performance affects career options. When we talk about some profession being underpaid or overpaid, we have to compare pay to other occupations that those currently employed could actually get.
And teachers' unions explicitly accept this. They compare teachers' salaries not to median salaries in the country, but to salaries for people with bachelors degrees, or masters degrees. But that ignores the wide range of academic performance represented by being a 'degree-holder'.
[0] https://drive.google.com/drive/folders/1ZOLlnHJ2OHZN3TuaVXky...
> I checked another district in California (Morgan Hill). Looking at all 2021 employees whose job title includes the word 'teacher', the median additional compensation is 21% of base salary.
Oddly enough, this is where I live. Random chance or stalking? luls. This is still a relatively well-compensated district compared to most other states or many other parts of California.
And ... even so.. 45% is not the same as 21%.
I have not checked your numbers to see whether 21% is correct.
Also, if you're including pay for additional duty (e.g. working over summer) in that 21%... that's not so great, given that you've accounted for that separately in your calculation.
> so I don't know whether those numbers include the present value of unfunded pension liabilities.
From the site you chose to link: "Total benefits" consists of the employer-paid cost of health, dental and vision medical insurance and retirement contributions only. The cost of benefits do not reflect monetary payments received by the employee but, instead, reflect the cost incurred by taxpayers associated with employer-provided health and retirement benefits.
> If so, the actual number is higher.
??? The actual number is higher because what we're comparing against includes NPV of benefits?
> When we talk about some profession being underpaid or overpaid, we have to compare pay to other occupations that those currently employed could actually get.
Yes, but using the SAT scores or other measures of academic performance as a proxy to determine career options is a faulty methodology. And even if it were a powerful determinant of career options, it would still be faulty, because willingness of people to engage in the work for the pay matters (e.g. you have to pay garbagemen more).
> They compare teachers' salaries not to median salaries in the country, but to salaries for people with bachelors degrees, or masters degrees.
Which would seem to be a reasonable measure. You haven't demonstrated teachers have lower SAT scores (you made a faulty indirect argument about this, too), they still have completed those degrees. Comparing them to a cohort that includes workers at McJobs is not fair.
I just want to clarify two things:
1. I didn't state clearly my point about the NPV of unfunded pension liabilities. AIUI, SFUSD contributes to teachers' pensions in two ways. The first is a contribution at the same time the teacher is paid their salary (which would show up on the 'total benefits' numbers we've been looking at). The second is contributions made after teachers are retired, when the contributions made beforehand aren't enough to pay for the defined benefits. When you look at teachers compensation today, it may account for the former, but probably not the latter, even though that has real value, it's paid by the employer, and it has real value for the teacher.
2. When you say 'workers at McJobs', I presume you mean those performing low-skill service jobs. AIUI[0], very few recent college graduates enter those jobs, so they're probably not a useful reference group.
[0] https://libertystreeteconomics.newyorkfed.org/2016/01/workin...
You are appearing to argue that comparing teacher salaries against all jobs is better justified than comparing against those requiring bachelor's degrees. Therefore, you are the one arguing for the inclusion of McJobs.
"And teachers' unions explicitly accept this. They compare teachers' salaries not to median salaries in the country, but to salaries for people with bachelors degrees, or masters degrees."
At the same time, we've shown on a metric you value (that I don't for this purpose-- SAT scores) that teachers likely outperform median college graduates. If we're going to compare comp, why should they not be compared to equivalently educated people?
https://cepa.stanford.edu/sites/default/files/Who%20Enters%2...
Especially for high status districts (like SFUSD), it looks like teachers are disproportionately drawn from the upper third of scorers.
The College Board defines 'college ready' as 1010 or above[1].
From the most recent College Board report[0] (page 4), the average score for those taking the SAT in their final year of high school is 1050. (I couldn't quickly find the corresponding number for 2010, but AIUI SAT scores are quite stable over time.)
From the paper you quoted, the average score for teachers is 1030 or 1064, depending on what you look at:
Between 1999 and 2010, however, average scores for those certified increases 27 points to a score of 1030 but the average score for entering teachers increases 65 points to a score of 1064.
[1] https://satsuite.collegeboard.org/k12-educators/about/unders...
No. There was a major test format change a few years ago, in addition to students becoming more qualified in general.
Reading scores moved from a mean of 500 to 529. Math scores have moved from a mean of 515 to 521. There's been a 35 point progression in the mean in that time.
> for those taking the SAT in their final year of high school
Yes, but a very large proportion of people just use their early junior year SAT score. So you're comparing a "final year of high school" vs "whatever year teachers took it in"
> From the paper you quoted, the average score for teachers is 1030 or 1064, depending on what you look at
You're missing the point of the paper. The people who actually progress to teach have a higher score than those who obtain a credential (many of whom do not successfully enter the teaching profession).
And, of course, note that teachers in desirable districts are substantially above the mean.