Amazon CEO Jassy Warns of AI's Unprecedented Adoption Speed, Education Shortfalls (yahoo.com) 44
Amazon CEO Andy Jassy has this week sounded the alarm on AI adoption speeds. Though self-described as an AI optimist, Jassy cautioned that this technological shift "may be quicker than other technology transitions in the past."
Jassy pointed directly to declining education quality as "one of the biggest problems" facing AI implementation, not the technology itself. He questioned whether schools are adequately preparing students for future tool use, including coding applications.
Jassy pointed directly to declining education quality as "one of the biggest problems" facing AI implementation, not the technology itself. He questioned whether schools are adequately preparing students for future tool use, including coding applications.
Clarify? (Score:1)
preparing students for future tool use
Is it the AI used as a tool by students, or the students used as tools for AI?
Either way, I feel those who push unwanted AI in everything are utter tools.
Coding vs Concept (Score:3, Insightful)
Too many schools are focused on coding without teaching computer science fundamentals.
That's leadingto coding which overlooks some things, and misses other things entirely.
"AI" (LLMs) can't fix that lack of education.
Oracle engineers took down 45 hospitals for 5 days during "routine maintenance".
That's not routine anything.
There was no peer review.
There was no backup.
There was no restore test.
But they probably got an A+ on coding and SQL.
Analogy:
If you don't teach the fundamentals you get people who know how to plug in a lamp, but have no idea how to wire up a receptacle, set a breaker panel, run conduit, build walls, lay foundation, set ingress/exgress for electric/water/sewer/gas.
So they have no business pretending they can do ANYTHING to build a house.
But they can plug in a lamp.
I've hired (and fired) people who start coding before defining the data structures.
DevOPS is for lazy people. "We can waste a lot more effort itterating than bothering to design right in the first place."
Ansible is an excuse for DevOPS people to say "We don't test our code, but we could if we wanted to."
Numerous other examples exist of tools that "enable" "coders" not to have to know crap except coding, and that
doesn't lend itself to long-term success.
Re: (Score:1)
According to whasisname at Microsoft, 30% of their code is AI generated so... we know why it's so spectacularly horrible. Also, that is so vague. 30% of what?
The race to the bottom (actually it's the middle) is well under way and Amazon is in hot pursuit.
Re: (Score:3)
"DevOPS is for lazy people. "We can waste a lot more effort itterating than bothering to design right in the first place."
Ansible is an excuse for DevOPS people to say "We don't test our code, but we could if we wanted to."
Numerous other examples exist of tools that "enable" "coders" not to have to know crap except coding, and that
doesn't lend itself to long-term success."
"Coders" don't know crap about coding either. That's what object-oriented programming ensured. Today's coder views his job as gluing to
Re: (Score:3)
Not really. We can argue, collegially, I hope. Data structures, like linked lists, queues, and stacks, aren't *procedural* code. Databases are also data structures, obviously, and also not procedural. If you have a system complex enough to *need* a database, you won't get v
They CAN'T know it all and I don't expect them to (Score:1)
Long-term, isn't this inevitable? They're expected to work at the top of that mile-high stack. They have to know the highest-abstraction stuff, and there's almost always going to be a point where there's simply no need to go deeper. The high-level language works and the programmer never needs to know how the hash or associative array or whatever-it's
Re: (Score:2)
DevOPS is for lazy people. "We can waste a lot more effort itterating than bothering to design right in the first place."
Ansible is an excuse for DevOPS people to say "We don't test our code, but we could if we wanted to."
This comment about DevOps and Ansible is worth unpacking a bit.
DevOps is a philosophy and a strategy for collaboration between development and operations, while Ansible is a tool for
automation. They are not substitutes for quality control, testing, or systems thinking. Effective DevOps requires a culture that
prioritizes reliability, safety, and design and it's up to an organization's leadership to promote that.
the old college system needs to be updated (Score:2)
At the very least an AA/AS should be part of an extended HS / 2 year community colleges (at HS costs to student)
4 colleges must take 100% of community colleges credits with no BS fees or bridge classes (unless they cover 100% of the cost of them)
No more BS like forced swim tests and if they want one it needs to be free.
Need more trades like education for tech jobs.
Jobs need to drop need degrees and don't use this an way to demand masters or higher for basic jobs.
Need more life long Education that does have
oh the irony (Score:2)
And now we can't find any more good data to train our AI because people are not smart anymore.
<womp womp>
Re: (Score:2)
Yep, funny how that works...
I do expect that if I want to do some FORTRAN coding after retirement, I will not lack in offers. At that tome (in 8 years or so), the deranged AI hype should be over as well and the actually not that great capabilities of LLMs will hopefully be seen as what they are.
"preparing students for future tool use" (Score:2)
I'm wondering what preparations he suggests? AI is rapidly moving into the entire coding domain, and as a result there won't be much payoff for investing in learning how to write basic software. I agree there will need to be a different educational paradigm for adjusting to this, but what would it be? Some form of prompt engineering? Any course of study you could imagine implementing today would likely be obsolete in a few years when the student would graduate.
Re: (Score:3)
AI is rapidly moving into the entire coding domain, and as a result there won't be much payoff for investing in learning how to write basic software.
In that case, prospective programmers should move directly to learning how to write complicated software, and just skip that first step.
Re: (Score:2)
Prospective programmers are going to be managing AI agents that do the coding. That's where this is going. I'm already spending most of my development time writing prompts and reviewing the results. Today I did have to write the signature of a new function I wanted to have implemented and some good comments describing what it should do. The AI then wrote 90% of the rest. I didn't mind it one bit.
Re: (Score:2)
Prospective programmers are going to be managing AI agents that do the coding. That's where this is going. I'm already spending most of my development time writing prompts and reviewing the results.
If you don't know how to write the code yourself, then you don't know how to verify that the software is doing it correctly. So either you learned how to program, and you're deluding yourself about how important that is to the process, or you didn't, and you're a delusional fuckup presiding over hallucinating software. Either way, nobody is going to be learning to use AI to write code correctly without knowing how to make the code work without AI.
Re: (Score:2)
>> nobody is going to be learning to use AI to write code correctly without knowing how to make the code work without AI
You obviously are poorly uninformed. People are doing this already, it is trivially easy. I'll give an example.
A couple of weeks ago I wanted some data from a publicly available resource in the cloud with a well documented API. I told the AI to pull down the data and store it in files with CSV format. It did that in under 5 minutes. Then I told it to create a website that would input
Re: "preparing students for future tool use" (Score:2)
"I had written no code whatsoever, but I could see with my eyes that the work had been done correctly."
You're so close to getting it, yet so far from thinking.
Re: (Score:2)
That all you got, boy? Clearly you have no experience with this tech.
Re: (Score:1)
Most of the problems I solve aren't as simple as setting up a basic API and doing auth. Generally I face problems that require editing large C++ codebases and adapting them to handle new functionality, while still preserving old functionality. I'm not adding much code, but I am editing a lot of it. These changes require deep domain knowledge.
In my experience, Copilot is worthless for this task. It is like an idiot just guessing at what should be ther
Re: (Score:2)
I hear complaints like that about Copilot from other people too. I've never used it. I suggest GPT-4.1 or Claude Sonnet 3.7, they run like the wind in my experience.
I bet editing large "codebases and adapting them to handle new functionality, while still preserving old functionality" is the job description of a lot of programmers these days. It sure has been my job many times. I think a good AI is well suited for this.
It can give you a detailed description of the existing code and generate a readme for you.
Re: (Score:2)
Complete code coverage is a useless metric.
if myCart.total() is more than 25 then
payment = myChart.total()
else
payment = chargeShipping() + myCart.total()
Complete code coverage means it'll test with a value above 25 and a value not above 25. It won't test with a null myCart. It won't understand that negative values shouldn't be valid and that check being missing should be considered a bug. It won't check if adding the shipping charge overflows the cart total.
Anyone promoting their 100% code coverage is shipping a buggy product. The fact that people who don't understand things like this will be pushing out AI cod
Re: (Score:2)
>> Complete code coverage is a useless metric
It simply means that all of the code has been executed, which is not useless and will expose certain types of bugs. Obviously you would want additional tests, and you can tell the AI to create those too.
"Combine code coverage with other testing strategies like unit, integration, and end-to-end tests."
https://muuktest.com/blog/what... [muuktest.com]
Re: (Score:2)
That all you got, boy? Clearly you have no experience with this tech.
My point is that if you didn't have experience, you couldn't look at what it had done and know that it was done correctly, not that the software can't do it correctly. Sometimes it can, sometimes it can't, and it takes a trained eye to know which has occurred. If you are as qualified as you say you are, it's because you got there the usual way.
Re: (Score:2)
>> look at what it had done and know that it was done correctly
I'm not disputing that. You do have to determine the AI did what you wanted. You also have to be able to tell it what you want in the right way, this is a skill that I am still learning. It is something that could be taught in school.
But its a moving target. It is much easier to get a good result than it was 8 months ago. Sometimes I am shocked at what it can do. The entire process may get quite a bit easier and require far less skill just
Re:"preparing students for future tool use" (Score:4, Informative)
AI is rapidly moving into the entire coding domain, and as a result there won't be much payoff for investing in learning how to write basic software.
What you said may be roughly equivalent to saying - back in the day - that "Very Large Scale Integration is rapidly taking over electronic hardware design, and as a result there won't be much payoff for investing in learning Ohm's law, semiconductor behaviours and properties, and the use of logic gates".
IANAP - I've only done a very small amount of programming, so I may be wrong. But I think that learning how to write basic software is analogous to learning the alphabet, having a small-but-growing vocabulary, and knowing basic grammar. People need to learn how to think about and evaluate the programming problems that AI solves, and how it might have tried to solve them. Also, since AI is trained on existing code, having a clue about that code is probably a good idea.
AI isn't always going to get it right, and I suspect that even figuring out how to prompt AI to correct its own mistakes will require some programming skills.
Re: (Score:2)
>> figuring out how to prompt AI to correct its own mistakes will require some programming skills
It does today for sure, but I find that the AI frequently will correct its mistakes if you just paste in the error messages. At times I have seen it run its own code and debug it automatically based on the print statements and stack traces it sees.
The pace of improvement on the AI's I've been using is pretty darn quick, a vast improvement over the past year. As it gets better you'll just need to have a rea
Re: (Score:2)
Thanks for the up-to-date info. I guess AI is just the latest example of how we have put layers of abstraction between ourselves and lower-level processes of all kinds. We've come a long way - from programming computers via toggle switches, to using natural language to tell the computer to program itself.
yes, of course (Score:1)
Of course a transition to AI will happen as quickly as billionaires can achieve it. It's profitable in its own right, but there's a lot more.
There is no faster way to enslave a population than to make them unemployed, hungry and homeless. AIs can be owned, people cannot (currently). Billionaires seek to own everything; AI can be owned. If AI can do everything, billionaires no longer need lower classes and do not have to negotiate for anything. And they will not. It is no surprise that Silicon Valley c
Re: (Score:1)
AIs can be owned, people cannot (currently).
False. Read the 13th amendment. We still have slave states and non-slave states. And even California is a slave state. We literally voted not to abolish slavery in California just recently.
Re: (Score:2)
You make no sense whatsoever. The Thirteenth Amendment in it's entirety:
Section 1
Neither slavery nor involuntary servitude, except as a punishment for crime whereof the party shall have been duly convicted, shall exist within the United States, or any place subject to their jurisdiction.
Section 2
Congress shall have power to enforce this article by appropriate legislation.
That sure seems to be abolishing all slavery to me. If I had to guess, you're a loon redefining slavery so you can call something you do
Re: (Score:2)
That sure seems to be abolishing all slavery to me.
That's because you can't read.
Tell us what this part means: except as a punishment for crime whereof the party shall have been duly convicted. Use small words so we know you understand them.
Re: (Score:2)
Prison slave labor is still legal and regularly practiced. Some states like to pretend they aren't slaves because they pay the inmates a few cents per hour.
Thank God AI writes itself (Score:1)
Holy Shit! Where would us knuckle-draggers if AI didn't write itself?
Which is it? (Score:1)
For the World Is Hollow and I Have Touched the Sky (Score:3)
This "warning" is completely disingenuous, of course. What is actually being said is that education as we know it is obsolete. Why bother learning anything when the AI can just solve it for us? As long as we have rudimentary literacy skills, everything else can be offloaded to the AI. I fear that ultimately we will all be trained to be utterly dependent on these amazing new tools. In a way, we are blessed to witness the birth of a new period in human history.
Are schools preparing kids.... No! (Score:3)
When it comes to AI, it's being demonized by plenty of teachers / schools / boards, but not for, generally, right or valid reasons. Last year, my younger daughter's teacher told me (paraphrased): "The math is so confusing, and it's difficult to understand the questions, sometimes.". That's where AI can come in, I'm not going to discuss if it's acceptable for a teacher to lack basic math skills, in this thread, I'm just going to state AI can be the new tutor.
The old saying: "The only bad question is one you don't ask", well, AI can probably answer it, and do a fairly reasonable job. Instead of having to raise your hand and look like an idiot, which is hard for a kid to do, they can ask AI, and that's what AI, from a school's perspective, can do amazingly well. What AI shouldn't be doing, is your homework, if the homework is worth doing. AI can review the work, AI can prompt the work, or help you explore lines of thought you didn't consider, but it shouldn't do that work. That's what schools need to promote AI as being great for.
Gun pointed at head: USE OUR AI (Score:1)
It's not like we have a choice.
Tech Mogul to English translation: (Score:3)
"We need more students who know how to make themselves obsolete, and be willing to make themselves obsolete."
Problem is Management (Score:2)
Re: (Score:2)
Re: (Score:2)
if you don't know how to do it, you can't manage an outside vendor to do it.
That isn't true. We all hire people to do things we don't know how to do. The problem comes when you don't even know what it is you need them to do.
Re: (Score:2)
Not a problem (Score:3)
Just wait until the hype collapses. And then it will just be one more small step that AI research has done and not in any way the "revolution" it currently gets sold as.
It drastically reduces classroom size. (Score:2)
All that said, I have used chatgpt to learn things. The beauty is that you can drill down as far as you want. It can mark your work, it can