This video argues that while prompt engineering is becoming the new coding and will make programming more accessible, it won't replace the need for core engineering principles. The history of coding shows a trend of higher-level languages abstracting away tedious low-level details to make programming faster and simpler. Prompt engineering follows this same pattern - it's an evolution of coding that makes instructing AI systems easier. However, prompt engineering alone is not enough. You still need to understand fundamental software engineering concepts like APIs, benchmarking, data structures, algorithms, etc. in order to write effective prompts and build real-world AI applications. Coding is just a tool; engineering is the skill of breaking down problems and designing the right solutions. As the video states, "what won't get replaced is the principles that govern data analytics, data science and engineering." The future lies in applying this engineering mindset to fields like drug discovery and climate change. Prompt engineering makes AI accessible to all, but core engineering expertise is still essential.
Thanks 4 the vid... I am Newbie... And want to make carrier switch... Into building, creating, enhance...ai Very interesting topic you posted! thx again.
As your 3-year follower, it's real strange that you use data scientist as your identity on RU-vid to produce videos and increase half a million followers and make money from RU-vid ads and sponsorships. But actually, you only worked as a data scientist for almost 1 year and quit your job to work as a full-time RU-vidr to make money! This 1 year of experience can't even secure a mid-level data scientist position in the current market, but you build a program to teach people how to learn AI and relevant knowledge. It's really ridiculous. I'm really curious about how much experience or knowledge in AI, machine learning, NLP, and computer science you really have, and how much you still remember. I wonder how much ML production experience you gained from your 1 year, only 1 year of experience. You are still a junior, but all your videos are teaching others how to work in the AI world and develop a career!
I imagine that future computer science courses will still depend heavily on the use of several coding languages in order for students to get a feel for how algorithms work and why some algorithms are better than others. Not teaching how computers work this way threatens to divide future generations into corporate morlocks and consumer/consumed eloi, with the former inevitably transforming (pun intended) into the latter.
Please make a distinction between writing a script to call an API and a complex huge program that uses QT or C++ and evolves over time together with the customer. Prompt engineering might be able to replace the first, but not the later. At least not in the mid-term future. Also good luck with debugging the mess once prompt engineering has coded >10000 lines of code. We are replacing the junior jobs with machine learning algorithm and in a few years when the old seniors leave, I wish you good luck finding someone who can work with the mess that is already there.
I reckon, that though Google dominates so much, and perhaps even AI productivity, there is more needed. A power of AI is that it gets to know you quiet fast, but we also need to be able to test and assess each AI service. Just as we need still to understand code, to see that it works well enough and is safe enough. If the Chinese take over a place, there seems little alternative, sadly true, but we still need to learn alternative possibilities, as history shows that the big power of authority is painfully negligent. It probably will be like that with AI enhanced services.
Sure, you should still learn to code in 2024 but finding a job might be more difficult than it's worth. Use the skill to start your own company....this could be video games, AI, whatever you want.
guys,I'm working on a coursework project and I'm looking into the quality of Gauth and how their grading systems work. Have you noticed any pop-ups or incentives while using them? I've only been downloading them for a little bit. i can well pay you for a short interview!!!help!
Thank you for posting this video. Your content is very helpful and it's helping me in my transition from serving in the military to now becoming a AI professional who lives in South Korea.
-Video: has a very clear question in the title -Tina Huang: "OK let me talk about history, technology, biology and whatnot. Basically saying all kinds of stuff except answering the question"
it's because she is business women who try to attract whith these question who make react people, but everybody know learning to code is still essential
@tinaHuang1 not quite sure i can agree that AI research began in the 1960s, given that there were others that laid the foundations to it such as the paper by Mcculoch and Pitts during the 1940s that set out the first logical steps to replicating the nervious system using mathematics, which then laid the grounds for Frank Rosemblatts Perceptron during the 1950s, that being a system that could identify perform a basic classification using computer vision, that being if an item was a triangle or a circle. Naturally this was very basic to todays standards but not quite as you lecture in your video (I say lecture as your podcast is a monologue of what you say without people discussing it as you talk with oposing views) so it becomes more like a broadcast and educational lecture
Coding is probably the least important aspect of software engineering these days. AI becomes a force multiplier rather than a replacement for developers. You still have to know how to apply what the AI generates. I can, however, hire one developer who is good at prompt engineering on top of the normal gamut of development skills instead of hiring 3-5 who can't leverage AI. The other thing AI can't replace is a software developer / engineer with domain expertise because those probably won't be commodity services for a while. If you add domain expertise to your AI model it becomes so large that burn rate with NVidia will not allow you to make a profit.
Wow. What a great video. Thank you for taking the time to produce it and make it available to us. You got a new subscriber; looking forward to bingeing the rest of your content when I can.
I don’t think prompt engineering will take over coding. But, it will surely help software engineers and other professionals like doctors, engineers, and lawyers to do their work better than ever before. It’s not a replacement, but a helpful tool.
She mentioned doing low-level systems and that reminded me of the little bit the equivalent class I took went over things like compilers and so on. For example: if you run sort(), the interpreter will choose the best options it can based on that call. It isn't just one primitive algo typically. Even more pronounced: you could go through the effort of putting an authenticator together, or the framework you're using might have a good built-in option that will work just as well and hurt a lot less. I haven't done much with any of these LLMs, but I got bored and made GPT give me the python code for a markov chain. Code looked great and ran just fine. I've also seen the same version muck up very simple problems that I remember solving in my intro class. I saw a post on reddit with a pretty good summary of this: "You should only be using these things if you don't need to." There's a shitton of people copy-pasting stuff that GPT, bard, etc. just vomit out and getting caught cause they didn't fucking read the output. A common concern I see is also GPT spitting out code that has license attached to it (which given how often it has given people stuff that was like the first result on google, bing, etc. that makes sense.), so the odds that this can fully supplant people is pretty low. Likewise the reason why unis still cover broad ideas like data structures and so on isn't *just* to take your money, it's also because if they taught you the most recent implementation of x it would be worthless. If they make you work with those more general approaches you can (hopefully) take that with you much longer. TL:DR, use it but don't rely on it. Don't swallow the fear kool aid before shit actually gets bad (Tbh I think I just wrote this cause of my own anxiety, been a bit uneasy about everything and needed to get my head right so I can go back to doing what I enjoy) ANYWAY, gg hf
Hello, just a little real example: I have this process that categorize customers in a Telecommunication company, 10M by now, a category is VIP/medium/low/debtors/others. We have 5 complete different core systems: billing, customer data, telecom transactional data, etc... The complete process involves these technologies: IBM Datastage (ETL), Oracle PL/SQL, Java, microservices, shell, workload automation, and some work in AWS lambdas. The idea, get the info from the cores, summarize, process and provide a category. My point, do you thing that in such process an AI could help to solve a business issue? My answer, NO. Do you think that the 'key user' can provide enough information to an AI to fabricate from scratch such process (forget about the stack)? for me.... NO. Do you think that an AI is capable to understand the business logic, including the 5 cores, and materialize a process as an application? haha, NO That's that, we are toying with AI. The main objective of all this is to automate and make repetitive tasks faster. But, the complexity of a business, game, auditory, modeling, critical system, still remains... For me, seriously, it is more feasible to replace a CEO w/AI to favor humanity for real.
Thank you for this video, Tina. I have been just a tad anxious about the effects of AI on current and future employment. This video helps me to make sense of it. Thank you.
The future is about devolving our ideas in a more comfortable way, "programming" and "software developer" is not going to disappear, it is just a transformation, instead of writing in paper we write code in files, the future would be probably writing an idea in a precise way so AI can understand and develop your idea, you should do some handwork even, but the future seems promising even though, we software engineers must adapt to the technologies.
I am sorry but given that this is what architects and product owners, or business analysis that analyse business processes and write down their specifications do - programming STILL had disappeared. All this spec work you define now is not something that is new, and it is NOT the domain of programmers now, and so why should it suddenly be?
I've stumbled across your videos three times now while searching for other things and every time I am impressed. Great content! (oh, and now I'm subscribed!)
Imagine if you needed a heart monitor and an LLM coded it and the Prompt Engineer did not catch a deadly code flaw. Imagine if that was your heart monitor? If we are coding without understanding what we are coding I don't see how we even begin to have any level of quality control.
@@darylallen2485 if humans are not following and understanding and keeping up on the tech how will they know that the products are safe and would we want to put our entire control into a machine that we would not in fact understand? Not sure.
You're very wise with the things you say. Thank you very much for the inspiration. I believe you have a lot of things to discover about the things you're doing on the internet, and it's on the right way to success. I hope you achieve great things out here.
The big wings need to be a bit more careful with their words. Typically they over sell and under preform. The need for a understanding of how it works is never going away.
We are lucky that there was no AI companies when UNIX and C were invented in Bell Labs or Linux was invented by Linus Torvalds or we would not have had these remarkable innovations
Thank you. I'll heed your advice.If things doesn't work out, what will I really lose? I'll take solace in knowing that I gave it my allPositive mindset and hard work will help me to achieve my goals and accomplish what I started one year ago💕😃Viel Erfolg
My mom does assembly coding for work. I dont know how she does it. I'm learning it in school right now and I have a new found appreciation for her because this shit is hard 😭.
@@melonpeaches9306 she's been in the industry for about 30-40 years making around 100-150k, she tells me it's much harder to break into assembly unless you've been doing it for a while, most of the people at her work place are generally older and people she's known already over the years working at different companies. Only assembly jobs are hard to come by and find unless you have connections in the industry already. It definitely feels like the people who are already good at it are getting the assembly jobs right now especially since it's a shrinking market. But don't let my words discourage you, eventually people retire and they still need people.
Excellent presentation. Prompt engineering, bio engineering, the history and future of programming -from a dynamic pharmacology and computer science co-major. Wicked smart, wicked inspiring. Thx Tina Huang!
As a biologist PhD neurobiologist (with large experience in experimental design, advanced statistics, SPSS, R, Python, SQL, excel, Tableau, learning Azure) aiming for a job as data scientist… I would love to have some advice in how to get a first job in a relevant industry.
Thanks I needed to hear this. Signed up for game programming in Unity but being forced to study things for sore purpose of learning what's going on under 'the hood'.
I have watched 75% of the video and still did not find the answer that the title of the video, taking so much time to make a simple point is really not a better use of time.
The answer is that yes you should still learn to code, because knowing how the program is built will make you a better engineer. You should also work on being more independent, more grateful, and less toxic.
I agree. This whole video is basically a data scientists view of software engineering and coding and data scientists are NOT software developers. So she did not really answer the question in her title and instead talked around the subject because, she is not a software developer.
Why wouldn't AI eventually become better at 'engineering'? Why wouldn't it be better at knowing what questions to ask, what architectures to implement etc.? What is the magical quality of the human brain that is unsurpassable for a computer and is it the same magical quality that just a couple of years ago people were claiming would make it impossible for a computer to produce beautiful images, videos, wite code, or solve mathematical problems?
4:11 not sure I can fully agree with you here, yes many dont know how to deal with memory management because those languages you mentioned are classed as high level programming languages as opposed to C which is low level. As such the programming language deals with memory allocation however Python which you mentioned, being used for mobile applications is not something I agree Python has been around since the very early 90s
I still believe programmers will have the most power to take hold of the best newest AI tech and monitize it first before it’s perfectly refined. Things and trust take time. It will happen but just the trust aspect won’t be there for years . I’ve seen coders make millions from AI projects because they were able to move faster than others. You’ll still have that advantage for many years
For those videos titled: "Should you still learng to code in {year}?", the answer is YES! Coding is essential in giving instruction to computers with zero ambiguities. AI applications are just a tool that have changed the landscape of coding, but will not replace coding.
Short answer...no, it will not replace. Long answer, AI will augment coding and developing. Just like anything else, those who are relying too heavily on typing into a keyboard need to up their game and start thinking like an actual developer.
Hi, the video was interesting, but do you know what made my day? You do the same kind of (more or less fitting) ST/Futurama memes I do on my T.clips channel 😅😂
I’m sure basic coding could be replaced or streamlined, but you still need people who have the knowledge and skills to find the best solutions. Companies have hired too many folks/managers who don’t have any technical skills nor the knowledge on the core tasks because they assume they “would know” without even knowing the basics.
For someone who identify herself as an engineer, data science, etc.. who has worked from assembler to high abstract languages, who has worked at meta and that is advocating for prompt engineering you made a huge mistake lady. You called react a programing language. It's so beautiful 😅 it's not even considered a framework. Anyone: Do you really want to know what prompt engineering is? Has it ever happened to you to go to the hardware store and ask for things you have no idea how is it called? Well consider those guys that know exactly what you need are the AI prompt. Now, guess what is going to happen if you try to get on this wave (get a seat, an opportunity) and all you are is a sales agent, a baker, a doctor. Yes that's correct you will probably hire someone else to do the job. So technically speaking, this video makes sense. Specially if you have the data. But don't be naive, not all databases are gold mines
I don't quite get how will prompt engineering can replace traditional coding. Your example where prompt engineering would be an additional level of abstraction is missing that the nature of the output and input changes. Here are somethings I would be concerned about : Consistency. 1.1 If you type in two prompts that have slightly different wording. The result can be vastly different, there are no guardrails to prevent variarions in results. 1.2 each time a new version of the ML model is updated to a new version. It does not guarantee backwards compatibility, imagine if java 11 is not mostly backwards compatible with java 7, think about the learning curve on each version. 1.3 ML models have randomness in LLM, but let's say we solved that by eliminating the randomness from the model, which may probably is doable. Let's say if you are able to somehow solve all these problems above. Then the machine learning model is no different from a traditional coding library that's deterministic. (With strict syntax ensure what you type is what you get.) So I just cannot think people would want to do engineering on top of non-deterministic behaviors. If people do actually do that, it is not called engineering, it is just delegating without checking the AI's work....
ARE core engineering skills still essential? I think AI makes possible a much higher level of "high level" than we even yet realize or care to admit. There are certainly indicators of it now, especially in the form of agents. I can imagine a very near future where AWS releases a core agentic tool that integrates ALL of its own products such that you can simply say to it, "Hey, AWS team, build me a website that does XYZ," and the AI assistant would be smart enough to ask follow-up questions to determine things like where the users will be located and what sorts of user and other data it would have to store in a database, as well as make rough predictions as to the resources required to meet traffic demands. The agent would know how to set up the database(s) based on the types of data involved and on questions of privacy and security. It would continuously monitor web traffic and automatically scale or suggest upgrades to the infrastructure. And it will generate beautiful websites because... well, we've already seen it do that. Of course, there would be hiccups along the way, but overcoming those is a simple matter of iteration and output monitoring. None of this would require anything remotely close to engineering expertise.
I doubt a 'coder' will improve much of qpu processings; thus I'd propose to focus at (architecture-)design to extend discoveries of quantum. There are ever more restrictions now. If I began coding today I'd choose sth. like RUST which itself is restrictive and add free code upon that can quickly be checked to disturb or not. A good idea might be to code an interface which switches amongst compilers/interpreters according to their advantages for a specific hardware architecture/setting.
I kept developing software projects without understanding the core concepts as a fresher people say you should do leetcode more, that is good but in an organization you need to build and engineer software …. Building personal projects needs to be on top…. even the person behind homebrew did not knew how to invert a binary tree but knew how to build things that are useful for millions of people to download and build software …. So fresher or Team Lead everyone has to keep learning designing and engineering software even if leetcode is a priority or not….
Thank you so much for this video Tina. This video was very helpful. It cleared all my doubts about prompt engineering. Also as a biology/medicine student, it really motivated me to work towards my field and learn more about drug engineering and life science engineering.
What he said that only coding is not going to bmget you job, you need to learn and specialize on other things as well. Because most of the time you just need to import libraries and write few lines of code. Imagine a ML engineer, how much codes he need to write right now, maximum 200 lines to trian and build a model That does not mean you will not need dedicated code writer, you will still need some but very few in numbers
according to the point of view of Jensen Huang that Tina shows us, it seems like a degree in science (e.g. biology) would be a better option in order to adquire domain knowlege because soon everybody would be a programmer.
Very interesting presentation and perspective. We definitively need to be able to see through the hype vs the real changes coming. Some skills are going to be obsoletes may be just a need to under the underlaying concept. Prompt engineering is just another layer of abstraction, but software engineering skills are still evergreen albeit in a different form.
The makers of Devin, Cognition labs have 3 job postings at the moment: Machine Learning Researcher, Software Engineer and another with the title "General Application". Would would a company with an AI Software engineer advertise for a Software Engineer role? One thing they know is the world is full of people you can fool.
i think y all need to stop posting these kind of videos like "IS CODING IS WORTH it in 2024!?" ooowh just stfu how abt that , there's people learning coding they don't want any disappointing shit like that, And yes i think coding is the Only Future also with other online specialties.
If I remember the internet, has been around since about the mid 50s, but publicly available around the early 80s, and the web, was implemented for public use around 1990 or so.
Tina, would you please recommend some books on computer science history? Thank you so much for your invaluable content! Love your videos, they're dense and light at the same time
How about reversing your learning methodology Learning/Implementing a Prompt-based script without learning the fundamentals of Assembly or C It seems that none of the "modern day engineers" today are willing to dig down to those very basic fundamentals?
She says coding in the title yet her entire video is about anything except coding. At the end of the day you have to give instructions to a computer about what it needs to do and to do that, you have to speak a language that the computer can understand, which is a programming language. Prompt engineering is NOT a programming language. It is basically an engine which translates human text into a programming language code. But someone needs to still make sure that this code is doing what it is designed to do, i.e. you have to test it, which often means you have to write testing code in a programming language, since you dont want to do that test manually. Then you want to also make sure that the code is properly structured so that it can be maintained and extended in the future. Code is not just something you write and throw away, it lives and evolves, often for years, and for that to work efficiently, it needs to be well structured and this is what experienced software developers, i.e. coders, do. So although it is true what she is saying that there are a lot of things around coding, such as requirement analysis and following a process to develop a piece of software is correct, you still need to develop that code and no AI I have seen so far can develop a well structured code base, which can solve a complex problem, and is written in such a way that it lives on for months or even years, slowly growing into thousands, if not tens of thousands lines of code. These things are not addressed by her because I suspect she is not a software developer but rather a data scientists, which is something else entirely.
For someone trying to break into Project Management in Software Development, how do you guys foresee the future? What should I focus my attention on to survive the next 10 years in this industry?
Yes. Human programming will still be useful for at least a few more decades. AI just isn't reliable enough at this time. You should really consider learning more about generative ai in your free time, though. It's pretty fun stuff. 👍
It's quite remarkable how much input you are able to apply in this one short video on a topic that would require three 2hr episodes. You are an expert on the subject, so it does help. There's a general misconception that Prompt engineering is just prompt. I think it's heading there, but I do agree to be ahead of the curve and to benefit from the current and emerging models, a comprehensive knowledge base will always be vital. But what the trend suggests is the endangerment of computer programming as an industry in itself. I thought it would take a few decades, but it seems it's going to take a lot shorter than that. It's already that many in the STEM industries are able to code as part of their day to day jobs, but there's evidence it will proliferate to all other industries and it's use will be as simple as just prompting (and not prompt engineering). The steps mentioned in the video to engineer a function, is going to be replaced by an intuitive AI that responds by giving optimal options (on cue). Google search already does that. But this isn't to say coding will be extinct, just not as highly sought as a stand-alone component on our resume.
