AI Disruption & Its Impact On Software Development Jobs

What will be the impact of AI disruption on the software development job market? Eric Evans & Dave Farley talk about the projections for future software engineering jobs and how Eric's experience working with artificial intelligence has led him to these conclusions.
This is a clip taken from Eric's FULL Engineering Room appearance, which you can listen to HERE ➡️ open.spotify.com/episode/2nBj...
-
🗣️ THE ENGINEERING ROOM PODCAST:
Apple - apple.co/43s2e0h
Spotify - spoti.fi/3VqZVIV
Amazon - amzn.to/43nkkRl
Audible - bit.ly/TERaudible
-
🙏The Engineering Room series is SPONSORED BY EQUAL EXPERTS
Equal Experts is a product software development consultancy with a network of over 1,000 experienced technology consultants globally. They increase the pace of innovation by using modern software engineering practices that embrace Continuous Delivery, Security, and Operability from the outset ➡️ bit.ly/3ASy8n0
-
#softwareengineer #artificialintelligence #ai

[@7rich79]

I think one thing to consider is how important the creation of efficient and flexible algorithms are on search, data processing, data analysis, compression and AI. However, we don't see an explosion in the number of mathematicians to work on these problems.

[@techsuvara]

This is actually a good point. The progress in software was mostly done through mathematics and logic thesis, not just necessarily computer science.

[@ContinuousDelivery]

Yet AI has already improved the performance of something as basic as sort, something that people had given up trying because we were sure that we already had done as good as was possible! oxsci.org/deepmind-sorting-algorithm-fastest-yet/#:~:text=Last%20month%2C%20DeepMind%2C%20an%20artificial,and%20efficiency%20for%20sorting%20tasks.

[@petersuvara]

@@ContinuousDelivery the devil is in the details. If you read the actual article, they did this : "To make the new sorting algorithm more usable for people, we reverse-engineered the algorithms and translated them into C++, one of the most popular coding languages that developers use. These algorithms are now available in the LLVM libc++ standard sorting library, used by millions of developers and companies around the world." - Reverse engineering an algorithm then working to get it to be more optimal? Looking at the optimisation the AI found, it looks like an obvious fix we used to do often when moving registers and checking min, max values, something commonly optimised. When we were working in ASM on the Gameboy and Gameboy Advanced, to squeeze the most out of these machines, we had to make some incredible optimisations. Such optimisations make sense when it's mission critical to get to a certain degree of performance. I feel like this is an attempt to find a suitable use case that has diminishing returns when it comes to training models. "If all you have is a hammer, everything looks like a nail" Certainly, there are specific usecases where these tools can find optimal solutions to certain algorithms and problems faster than a human, so in this case, I believe just like any optimisation problem, neural nets are perfect for this. But it's not replacing anyones job at this stage, it's like you said, an effective addendum to our toolset as developers who solve problems. Also, I think we should be mindful of the cost of training models (money, storage and energy) vs the saving they provide in efficiency in other areas. In this case, it becomes a business decision, what is the best thing you can train a neural net to do that saves the most amount of time and energy.

[@et379]

This is just advertising. They did not create a new sorting algorithm that was faster. The 70% boost was gained by optimizing the underlying asembly Code of the existing sorting algorithm in libc++, by changing 2 assembly instructions. While that is impressive and amazing, the title and the claims are misleading. A more accurate description would be to say that it optimized an existing LLVM libc++ algorithm to perform 70% faster for a specific set of CPUs and specific integer data types. Still great, but far from creating a completely new sorting algorithm.

[@j.oliveira]

Nice vid. I have that Surf Arrakis shirt as well! 🤣

[@johnjakson444]

Utter bollox from people that don't seem to know what AI code writing can actually do.There are elementary coding exampls that contain barely a single idea that represents a coding problem and AI often fails these simple tasks. A real software development project has 1000s of complex ideas and architecture that AI will never gets its tiny thinking skills around. The only job growth will be in the hyperware marketeer nonsense artists, we have seen this happen every few decades.

[@jasonkitchen968]

What if humans stop writing code? What will the AI train itself on? Seems like a path to stagnation.

[@petersuvara]

Already happening with imaging, the quality of AI generated images has been decreasing as the data is trained on AI generated images. Kind of like a cannibalisation.

[@vitalyl1327]

LLMs write a much better code with ReAct, in languages they were not trained on (ideally, languages that did not even exist in the training set).

[@Bozebo]

I'm seeing plenty of ways ML can be useful in systems but not worried about it replacing people in designing systems any time soon.

[@gammalgris2497]

The level of training and calibration with "AI" reaches new levels. Artificial neural networks are useful tools to approximate unknown functions or computational heavy functions. It may kill your process and productivity if it increases your manual testing. OpenAI & others have crowdsourced that aspect and people pay for testing such services. What I want least is a monopoly on black boxes. Black boxes that are echo chambers of what is popular.

[@techsuvara]

Training models is the biggest hurdle, the amount of energy it took to create the GPT-4 model was enough to power a town, for... check this, ONE YEAR! All it can do is condense basic information for reference, it's a good reference manual, but that's about it.

[@MykhayloS]

Do you have numbers of how that compares to the energy consumed by Netflix infrastructure to serve the final part of Hunger Games to global audience? Is that somehow comparable?

[@matswessling6600]

seems like this "risk" is based on an misunderstanding on what software development is.

[@Imscottirl]

We are literally on the precipice of a software development boom like we have never seen and all the old guard can talk about is doom and gloom.

[@pablopelaezgallego4806]

*Boom for the ones who are ready to get into it.

[@Immudzen]

We have also been on the precipice of self-driving cars for more than 10 years and progress has nearly stopped. It was easy to get to about 95% but getting beyond that has proven to be insanely hard. These LLMs are already hitting the same point. In the hands of a skilled coder they can assist but they don't understand. Most of the code they generate is wrong to various degrees and when an LLM is questioned about the code it is pretty clear it has no understanding of what is being asked. Remember LLMs are probability models where they predict the next most likely word. They don't actually understand the question being asked. I think they will get better as assistants but won't be able to replace. There are some companies that already tried to do all the work with LLMs and no coders and they failed. I think they will continue to fail.

[@Imscottirl]

@luke5100as I replied in the other comment, the video is excellent. It’s the marketing that’s getting old.

[@Immudzen]

@luke5100 I don't dispute it is a timesaver however at the same time it is also very stupid. It doesn't understand what it is doing and no amount of prompt writing changes that. There are quite a few research papers on this already. You are correct that the more narrowly the scope of something is the more likely it is to succeed but that also very clearly moves it into assistant and not something that can be used without programmers. I would say that on scientfic and engineering code if I have it write a 5-10 loc function it will usually get about 80% or so of it correct but you better know how to write that function yourself or you are never going to be able to correct it.

[@vanivari359]

yeah, obviously we are good for 10 years, the next 10 years will be great for talented software developers because those tools remove countless hours of boilerplate code and just boring programming - i can do a job now that previously was done by a full team in India and i had to review and fix their code (we have 200.000 employees in India for some reason and most of them are really not good software developers, they just had a Java training). My job will be safe until the last minute. But AI is already good enough to produce results like a lot of junior developers, but in seconds vs weeks. And instead of writing review comments and spending hours in peer sessions, i can fix it in 5 seconds by just saying it. And once AI understood how it should generate test object factories, it will do it exactly that way - for the whole team or company. With a better integration into the tools and legal issues solved, this could already replace 30-50% of our classic developers. Yeah, AI makes errors, but have you seen how many errors humans do? How many humans have barely understood what they are supposed to build? Millions over millions of developers out there are not the top 20%. But how are people learning our trait, if AI is better and cheaper than most junior developers? And what is our society doing, if 20, 30, 40 or 50% of high paid jobs disappear in 10 or 20 years? The ones who pay the majority of taxes and consume the most products and services? Who needs to build software, if there is nobody left who can effort the products? The first manufacturing robots looked silly and they turned an industry upside down. But that is nothing compared to the potential impact of AI.

[@techsuvara]

I did a video on a new AI based language. I mean, there's more to code and more things to build than ever before. I wouldn't be phased. Stay hungry, stay foolish, stay curious and make magic happen...

[@zvuho]

AI will not take your job, people that can use it will.

[@Imscottirl]

I think you would have more clicks by canning the fearful messaging, and offering hope instead. Lots of devs are sad and depressed; they are looking for anything to make them feel better about their future. They’re sick and tired of being manipulated by fear.

[@michelmagix]

Ask a politician then. I think to face the reality is way more helpful to see future potential for new jobs than sweet talking the whole situation. A.I. is a new species. Hence imho we should all share facts and learn from practical experiences on the go. Then hope has a honest base to flourish.

[@o.fm.a5573]

@@michelmagix do you think fearmongering is being honest about highly skilled people?

[@Imscottirl]

@luke5100I’m not criticizing the video itself. That’s great (as always). It’s the marketing of the video that is somewhat fear mongering.

[@ordinarygg]

Everyone talking about this, but still no impact on software development. All prompt results are super-bad still. All Hype will gone soon

[@ballonura]

so agree

[@josediaz5663]

Although you are right with the current state, it's s a continuously evolving technology so it is definitely a possibility

[@stocothedude]

Lot's of software engineering/programming jobs will go away but also new jobs will arise.

[@michelmagix]

Yes. And if not, we should all think for ourself and as a collective what should we do with our time to create value other than 'just' doing our 'work'

[@matrixnorm6672]

IT will implode in 20 years. Imagine clueless 40yo zoomers prompting clueless AI. 😂

[@ContinuousDelivery]

Pretty sure that in 40 years time, the AIs will be "prompting" themselves!

[@GROW_YOUTUBE_VIEWS_m043]

This sound is relaxing

[@michaelrstover]

Imagine not understanding how putting the 'A' before the "GI" changes everything. I mean, what? How does such a person imagine we get to AGI and then somehow just stop there? Firstly, the same folks who achieved the AGI are going back to work the next day to achive AGI V2. And so on. And they'll have a new tool to help them, can AGI! And second, AGI will also be put to work to develop AGI V2, and they'll be tireless. They'll clearly be super human in various ways we can already see in LLMs right now, at the very least involving a breadth of knowledge recall no human can match right now. I don't know, this interview seems to disqualify Mr Evans and his opinions, from where I sit.

[@samcalder6946]

At best, AI allows you to accelerate mediocrity. The reason people see it as such a threat is because 90% of the workforce are actually just trying to produce something that is "good enough" as fast as possible. Until we get quantum computing at scale, there will not be enough compute power available for AI to produce anything that requires creativity, innovation or critical thought.

[@samcalder6946]

The output is simply a derivation of the input (in most cases, simply the most common input in the training data). Potentially useful for boilerplate code. Maybe useful when you're learning to get up to that "mediocre" standard, as long as you do not trust it enough to be performance, canonical, bug-free, or even compileable without further verification. Largely useless when trying to do anything new or novel. As an example, perhaps "data entry staff" may have something to fear... They may need to retrain as reviewers of the AI's work, rather than doing the work directly. In reality I suspect companies are more likely to scale up/out their data processing as a result, rather than scaling down their workforce.