[SpeedBird]

Originally Posted by ValarMorghulis

for now.

Wait a few more years. AI can be eye opening.

Why few years? Give it many years. How much every time you want. But it can never create new things. AI by definition is echoing the existing data in a modified form. If you have a tech that is creating new things, its no longer AI. It is Natural intelligence which we humans have and a machine can never match the creativity of humans. To achieve those levels, we need to understand the working of human brain, which till date is unachieved. Keep in mind the definition and concept of AI and you will automatically know what all it will replace, what all it will leave and what all new opportunities it will generate!

Originally Posted by Janus

Only a handful, in my 15 years of experience, are willing to migrate away, simplify, minimalize and move on to the next thing.

I can clearly see that you are not one of the technical software engineer/coder out there who knows how the industry works.

Name me a few companies who are willingly ready to give their proprietary data to the AI gaints rather than developing their own in-house AI. So when in-house AI solutions are developed, it can play around only with the dataset within the company. With most of the code being undocumented and uncommented, and layered to such an extent that the dependency code itself is long gone from the visibility but is there in production, with this level of complications wherein only the humans who had worked in that code previously knows what's happening, and on top of that each company having its own standards and approaches, how will this advanced AI tackle the incoming bugs for those applications? Forget about even creating new things here, solving existing thigs itself is far from reach from AI.

The max it can do and is doing currently is helping the developers in writing code by auto filling the redundant code. This is helpful for software engineers rather than replacing them.

Remember dotcom bubble? Yeah its a repeat of the same.

So give it few years, and you will see the repeat of dotcom bubble burst in the form of AI bubble burst.

[ValarMorghulis]

Originally Posted by SpeedBird

...Give it many years. How much every time you want. But it can never create new things. AI by definition is echoing the existing data in a modified form. If you have a tech that is creating new things, its no longer AI. It is Natural intelligence which we humans have and a machine can never match the creativity of humans. To achieve those levels, we need to understand the working of human brain, which till date is unachieved. Keep in mind the definition and concept of AI and you will automatically know what all it will replace, what all it will leave and what all new opportunities it will generate!

Artificial Super Intelligence (ASI) or Artificial General Intelligence (AGI) is the next step for AI.

Here is a very nice 2015 article which easily summarized AI and its future. Quite a many scientists predicted 2040 as the year when ASI/ AGI will be reached. Check it out. A nice article irrespective of our differing views

Link

[warrioraks]

Programming jobs are going to change significantly over next few years. Coders who use AI will replace the ones who don't want to. Won't happen instantly but give it some time.
AI is going to be a gamechanger like internet and my bet is that it will evolve very much in-line with Amara's law.

[Jeroen]

Originally Posted by SpeedBird

With most of the code being undocumented and uncommented, and layered to such an extent that the dependency code itself is long gone from the visibility but is there in production, with this level of complications wherein only the humans who had worked in that code previously knows what's happening,.

Undocumented, uncommented code? Is that the new standard of coding?

It has been a while since a ran software development program. Nothing too big, medium size, in those days about 1,5-2,5 million man hours, which we would typically out in 12-18 months or so. Part of our quality requirement was the way all our programmers documented everything. We had very stringent standards and without all relevant documentation in place, checked and certified, that particular software would not be allowed to go into production. It’s an integral part of software development. Or it should be, in my opinion.

From a costing point of view we had norms of how much time was spend on proper documentation of various parts of the software development. It was a sizeable percentage of the total man hours required.

Later in life, I was part of various due diligence teams that investigated companies we might wanted to acquire. I have had the pleasure of looking under the hood, so to speak, of quite a few (software) product companies around the world. Not having detailed documentation (in proper English) was a huge red flag to us.

[warrioraks]

Originally Posted by SpeedBird

Why few years? Give it many years. How much every time you want. But it can never create new things. AI by definition is echoing the existing data in a modified form.

Hallucination is one of the biggest challenge GenAI implementations are facing right now. These guys and gals productionizing the stuff would have really preferred that LLMs work like you mention (just echo the bloody data).

Originally Posted by SpeedBird

Name me a few companies who are willingly ready to give their proprietary data to the AI gaints rather than developing their own in-house AI.

Any and every company which is using ChatGPT, Github, Microsoft Copilot, OpenAI models and wrapper services on top of these in one form or other. Which practically is 90% of the organizations out there.

[DigitalOne]

Originally Posted by SpeedBird

Why few years? Give it many years. How much every time you want. But it can never create new things. AI by definition is echoing the existing data in a modified form.

I am not sure what your definition of 'create' is, but as of today AI is creating text, images, videos, music.

Where humans are underestimating AI is in comparing (current) AI with the best of humans in that field. An AI's output may never reach the level of a Shakespeare, a John Grisham, a Raja Ravi Varma, or an AR Rahman. But in most of the creations, AI is currently better than an average human. Average being the keyword here.

Same will be with programming. AI will replace average joe programmers in a few years.

Not so OT: It is well known humans overestimate themselves. (I believe that I am among the top 5% writers in team-bhp ). This effect is known as the Dunning-Kruger effect. When reality hits, they are often shocked.

[SpeedBird]

Originally Posted by ValarMorghulis

Artificial Super Intelligence (ASI) or Artificial General Intelligence (AGI) is the next step for AI.

Here is a very nice 2015 article which easily summarized AI and its future.

I remember the hype of Devin, the software engineer AI bot lurking around for few weeks, which scared all the so called software engineers who are surviving out there "politically". Even then, I don't see Devin snatching away the job of such engineers, forget about genuine software engineering jobs.

The same type of articles were lurking around in the then newspapers of the late 1990s when www was in boom. Yet we still see people buying newspapers. The ones who were majorly hit with dotcom bubble break were those people who considered themselves "web experts". Sounds like a deja vu eh?

Let me know your thoughts ValarMorghulis. ("Yes, all mem must die, but we are not men". So, you know some things won't ever die!)

Originally Posted by Jeroen

Undocumented, uncommented code? Is that the new standard of coding?

Thats how it always was! Documentation is a myth, a process which true logical programming software engineering minds hate to do. Its a tendency of humans to have a liking to a particular job role. Those who liked documentation were never good at logic and reasoning. The only codes and programs which AI can generate are the ones which are already there. If it were actually the hype it is currently creating in the market, all software engineering jobs would have been vanished by now. Slowly the companies are also realising that it is a bubble. A concept called model collapse is a real thing. By the time the "AI investors" realise their mistake, model collapse, along with many other factors would have already caused the bubble burst

Originally Posted by warrioraks

Any and every company which is using ChatGPT, Github, Microsoft Copilot, OpenAI models and wrapper services on top of these in one form or other. Which practically is 90% of the organizations out there.

Partly agree with you. Yes they are using the AI tools at the fullest extent. But the data being fed into these tools are completely proprietary data. Just how private github works. You are using github, but GitHub as a company doesn't know what data you are feeding it. There are licenses and laws which make sure that the privacy of company 's data is maintained. If it weren't the case, we would have already faced a digital disaster by now. The only data which gets sold without consent are the user data (no matter what your host app/website says) and the AI models can only feed on those. Unfortunately, they aren't reliable to an extent which will end up replacing human intervention

Originally Posted by DigitalOne

I am not sure what your definition of 'create' is, but as of today AI is creating text, images, videos, music.

"Creating" is an exclusive skill of any living thing. Ofcourse AI can create things in the way you mentioned. But what you think as "creating", is actually a blend of 1000s of parameters, in the way which feels a new thing to humans because there is no possible way for a human to know each and every website and data there is today.

But this is not the definition of "creating" in the software world. Creating here refers to the problem solving skills and solution oriented approach which would help us come up with new innovations based on real world requirements. This is one of the key aspects the "AI startup companies" miss out on. It is being realised by the investors, but the process will take few years. Sad part is that the new grads who are behind this "new tech trend", and only behind this and neglecting all other important subjects related to problem solving and logical programming, will end up loosing their career and with a huge debt.

Note from Support: Posts merged. Please use the edit / multiquote functionality instead of back to back posts within 30 mins on the same thread.

[warrioraks]

I am going to pick some snippets from your last two posts to continue this conversation, since I really like to discuss applied side of AI.

Originally Posted by SpeedBird

Name me a few companies who are willingly ready to give their proprietary data to the AI gaints rather than developing their own in-house AI. So when in-house AI solutions are developed, it can play around only with the dataset within the company.

Most non-tech companies will never be in a position to compete with bigtech on building these LLMs. They will have no choice but to provide their proprietary data beyond the four walls, just like it happened with cloud and other SaaS services.

Originally Posted by SpeedBird

Remember dotcom bubble? Yeah its a repeat of the same.

So give it few years, and you will see the repeat of dotcom bubble burst in the form of AI bubble burst.

Tech industry in general has been in overheated territory since post-Covid madness. Software developers with few years experience are earning crazy salaries. You can argue that AI is hyped up, but the current bubble in the industry has nothing to do with recent developments in AI. The software engineering bubble has been brewing for way longer.

Originally Posted by SpeedBird

I remember the hype of Devin, the software engineer AI bot lurking around for few weeks, which scared all the so called software engineers who are surviving out there "politically". Even then, I don't see Devin snatching away the job of such engineers, forget about genuine software engineering jobs.

The same type of articles were lurking around in the then newspapers of the late 1990s when www was in boom. Yet we still see people buying newspapers. The ones who were majorly hit with dotcom bubble break were those people who considered themselves "web experts". Sounds like a deja vu eh?

Devin or Replit or Cursor is NOT going to take away jobs. But it will definitely burst the software engineer bubble where the average folks will no longer be able to job-hop every few months and get crazy hikes. Those days are gone. In any industry, the topmost talent always commands a premium. This will continue to be the case software industry as well.

Originally Posted by SpeedBird

Partly agree with you. Yes they are using the AI tools at the fullest extent. But the data being fed into these tools are completely proprietary data. Just how private github works. You are using github, but GitHub as a company doesn't know what data you are feeding it. There are licenses and laws which make sure that the privacy of company 's data is maintained. If it weren't the case, we would have already faced a digital disaster by now. The only data which gets sold without consent are the user data (no matter what your host app/website says) and the AI models can only feed on those. Unfortunately, they aren't reliable to an extent which will end up replacing human intervention

Proprietary data going out of the four walls of the company is not the same as proprietary data being used by 3rd parties without consent. In this case, companies will be forced to feed proprietary data into these models for the purpose of inferencing. However through contractual clauses, they can always restrict BigTech from using proprietary data for model trainings. Guess we are talking the same thing here.

Originally Posted by SpeedBird

"Creating" is an exclusive skill of any living thing.

What would be the basis of this statement?

Originally Posted by SpeedBird

Ofcourse AI can create things in the way you mentioned. But what you think as "creating", is actually a blend of 1000s of parameters, in the way which feels a new thing to humans because there is no possible way for a human to know each and every website and data there is today.

Repeating my point from the previous post - hallucinations (and therefore creating new stuff) is a feature of these GenAI models. People are using it to create new songs, deepfakes, images of Trump and Harris on the wedding aisle, and what not. Although I don't condone a lot of these things, the fact is machines are able to create stuff which was not part of their training dataset.

[Tensor]

My logical and educated guess in this whole GenAI and software engineers loosing job.

This is a long read; anyone’s welcome to skip. If you don’t know any of terms used google it or ask an AI.

Firstly, let’s talk about what it takes to replace most of software engineers i.e. 60-70% with AI (no complete replacement is simply not possible even with ASI forget AGI we need some people to overlook/use AI and to maintain existing code infrastructure).

The LLM must be able to read into and write into large code bases like really large ones since most of the time, we’ll be just tweaking/refactoring existing code bases and not writing from scratch for we need really large context windows like I’m taking 2M + currently, most LLMs have either 128k, 200k or 500k which is really challenging and is limiting, but there’s an exception. Gemini Advance already has 2M context window, but is quite dumb and cannot match SOTA LLMs.

Secondly we need AI to not just spit code out instantly, but to architect the software first and then, logically think further steps to take like which platform to use, what programming paradigm and whether to containarize it or not etc. which is again being solved by another LLM o1 by OpenAI but the issue is it’s context window is 128k as of now which is a limiting factor.



Then, we need an LLM which writes excellent code without many errors and edge cases handled well. Sonnet 3.5 does it you can literally ask it to make Tetris (Tetris as a game is fairly complex not like snake) and it makes it one shot. Yes, that’s it everything works from animation to game logic to control it’s superb or take a piece of paper draw a website layout it gives a fully functioning front end code which looks exactly like you wanted (obviously you must tweak some things which can be done in 1-2 shots later) and then it’s equally good in backend as well recently without any fuss, I was able create a login with google, added recaptcha to a website or embed google maps straight to a website in like 2-3 shot prompts no LLM could ever do it before this quickly and not just that creating fully functioning clones of WhatsApp, YouTube etc are a breeze and can be made in 60 mins (I’m taking full stack/ fully functioning applications not a dummy front end copy)
These things were never before possible.

There are a plethora of videos on X and YouTube of people building amazing software products using cursor and an API of some LLM. It’s really worth checking out!



So all we need is reasoning capabilities of o1 + coding skills of Sonnet 3.5 + context window of Gemini advanced. All of these exist. It’s just that they exist separately and not as a whole. Only a matter of 18-20 months before all these things will be enabled in a single LLM which will be even better in coding tasks than Sonnet 3.5.

Some argue that privacy of proprietary code, well, usage of any leading LLMs via enterprise plans or APIs won’t be used as training data or just use open source LLM like llama 3.1 405b and fine tune with codebase and run locally without internet . Also much advanced and capable models will arrive in future for free.



Finally, people say it’s expensive to run LLMs. Well, not for long usage of complete renewable energy is a one time investment. Just see how chips are getting faster and efficient. Look up hopper vs Blackwell and you can see Blackwell is 2000% faster (most LLMs were trained on hopper H100s). Fab designs are still expected to have exponential improvements amd may still follow Moore’s law.





Add to that, in less than 20 months we have Sonnet 3.5 from ChatGPT (GPT 3.5 the dumb one). Sonnet 3.5 steamrolls any college graduate with a computer science degree who either spend the 4 years of college rote learning DSA or hop on the MERN stack trend instead of being flexible to any tech stack and even mid level devs with 5 yoe already and it only keeps getting better. The flagship model, Opus 3.5 comes this year.

Saas, IT Service, Tech startups with minimal users showing only MVP as an industry, will only decline from here and will consequently lay off a massive part of their software engineers.



Credit/Read more:
https://x.com/carrynointerest/status...637631356?s=46

Klarna is an example of how they ditched Salesforce and built a simpler in-house software using AI. Expect more businesses to do this.

Source: https://www.inc.com/sam-blum/klarna-...m-with-ai.html

Essentially, most of the billion dollar IT exports are just GPT tokens. In the future with costs a few cents/million tokens.



Anyone who is delusional that AI can’t replace them, are just coping and won’t help them. Teams will shrink. 5 people can achieve the work of 15 easily and those 10 get displaced. Moreover, most devs in India in service sector are not highly skilled. If you’re experienced, try switching to a higher package or if you’re a fresher, try getting a decent package with no bonds. You’ll realise the current state of market.

If you still think it’s high interest rates (companies are posting record profits YoY, why do they need debt to fund hiring) or over hiring in pandemic, well good luck.

Software production will have a massive boom with less people pumping out more software, while software devs will have a hard time finding jobs and will most likely, be paid less, expected to work more and will have to face many rounds of interviews since they’ll be competing with lakhs of other experienced devs and will be in PIP if they don’t work for X number of hours stipulated by the company.

All things considered, it likely takes 36-48 months more for most of the entry and mid level software devs to become redundant and makes AI cheaper and more intelligent. Most of tier 2, 3 B.Tech CSE grads will have a hard time getting jobs and potentially their degree might even be useless since software engineering, as a whole, gets commoditised to everyone and with a few prompts, building large scale apps will be available to all.

Obviously, there won’t be a sudden layoff situation where everyone gets laid off simultaneously. Rather it’ll be a slow decline with less hiring and more periodic firing of people in chunks over the decade. The current trend of slump in IT jobs would very likely continue.

If I’m wrong, please educate me with technical and logical arguments.

[Samurai]

Originally Posted by Tensor

If I’m wrong please educate me with technical and logical arguments.

Can you add your professional experience to your post. What is your industry domain and whether you are in services or products, etc.

That will help BHPians to respond to you at the right wavelength.

[Tensor]

Originally Posted by Samurai

Can you add your professional experience to your post. What is your industry domain and whether you are in services or products, etc.

That will help BHPians to respond to you at the right wavelength.

I have done my engineering in computer science and have been in this GenAI space since the era GPT-2 when it used to dumb and generate repetitive words , me and my friends used to build wrappers on top of these LLMs for fun in college days, so I’m quite familiar with the transformer architecture.
Currently I run a business which is not completely related to this industry.
I have a few friends in service sector and startup founder with whom I chat and share thoughts.

Would love to hear from more experienced folks serving in product and service sectors what they think about this

[Samurai]

Originally Posted by Tensor

me and my friends used to build wrappers on top of these LLMs for fun in college days, so I’m quite familiar with the transformer architecture.

Oh, when did you graduate?

Regarding your friends, are in the similar age bracket or do they have extensive experience in IT industry?

Please bear with me, I am asking these questions after reading your post. I am really curious about your vantage point. After all, you are making very serious predictions about the IT industry.

[Tensor]

Originally Posted by Samurai

Oh, when did you graduate?

Regarding your friends, are in the similar age bracket or do they have extensive experience in IT industry?

Please bear with me, I am asking these questions after reading your post. I am really curious about your vantage point. After all, you are making very serious predictions about the IT industry.

Not too long ago, 2021 I know I have really low experience in typical software but mostly in GenAI & transformers/diffusion models as they really took off in 2020 when GPT 3 was launched (predecessor to chatgpt).

yes friends were classmates so same age bracket.

This whole idea of approximate 36 months came when I saw a podcast of Emad Mostaque founder of stability ai who said in a podcast that an uber like app could be build by LLMs in one year from now hence the calculated guess

Podcast:

https://youtu.be/VeaT6ucXuMI?si=eWVPKF2749HVhJWz

What most people dont observe is the exponential progress of compute they say models wont improve but for models to think hard like o1 they need a lot of inference, with chips getting better according to moore's law we can confidently expect better LLMs its not a matter of data, just compute and nothing else.
After hearing this and from few reddit comments came to this conclusion.

[Samurai]

Originally Posted by Tensor

Not too long ago, 2021 I know I have really low experience in typical software but mostly in GenAI & transformers/diffusion models as they really took off in 2020 when GPT 3 was launched (predecessor to chatgpt).

Ok, thanks for sharing further information.

Since we have extensively and frequently discussed this topic (AI taking away IT jobs) in this thread, I have moved the posts here.

We often hear AI pioneers like Sam Altman, etc., making similar predictions as you. However, they simply don't address the elephant in the room, the customer expectations in IT services.

As far as my own views are concerned, I summarized it in view posts starting here.

[SpeedBird]

Originally Posted by warrioraks

Most non-tech companies will never be in a position to compete with bigtech on building these LLMs. They will have no choice but to provide their proprietary data beyond the four walls, just like it happened with cloud and other SaaS services.

Agree, that's how AI is effective in generating good code. But its not the best out there. I am not talking here about small and medium level IT softwares and tools. I will talk about this down the line

Originally Posted by warrioraks

Tech industry in general has been in overheated territory since post-Covid madness. Software developers with few years experience are earning crazy salaries. You can argue that AI is hyped up, but the current bubble in the industry has nothing to do with recent developments in AI. The software engineering bubble has been brewing for way longer.

Software engineering bubble is there from a long time. I completely agree on the fact that there are people who are surviving purely based on politics and job jumping skills. I did mention this in my previous reply and yes its specifically the time for them to go out of this software world.

Originally Posted by warrioraks

Devin or Replit or Cursor is NOT going to take away jobs. But it will definitely burst the software engineer bubble where the average folks will no longer be able to job-hop every few months and get crazy hikes. Those days are gone. In any industry, the topmost talent always commands a premium. This will continue to be the case software industry as well.

As stated just above, completely agree with this. But in parallel, the hype of "AI engineers" also is not something which will last long and the most toxic thing is that these "AI engineers" aren't the true software engineers. Unlike what others think, they belong to this "average folks" you mentioned. They don't know the basics and fundamentals of engineering. I have my friends and colleagues who don't know anything about software development life cycles and how a company operates and going to do MS in the field of AI by taking crores of loan. When they come back, the hype of AI would be gone and they would need to switch back or rather adapt to the new normals after the AI bubble burst, which would be near impossible for someone who has studied and believed only in AI and don't know how to do things traditionally. (One of the best analogy I can give is: trying to driving a manual car when you are used to driving automatic cars. Fundamental knowledge is always important in any field.)

Originally Posted by warrioraks

Proprietary data going out of the four walls of the company is not the same as proprietary data being used by 3rd parties without consent. In this case, companies will be forced to feed proprietary data into these models for the purpose of inferencing. However through contractual clauses, they can always restrict BigTech from using proprietary data for model trainings. Guess we are talking the same thing here.

Agree, its pretty much the same. BigTechs escape, in the same way Amazon, cisco, eBay etc survived the dotcom bubble. Once the proprietary data of startups are in AI, it will ruin the company. Forget about if it improves the model or not (a different topic which depends on the talent of the engineers working in those startups), but once their data is out, it can be reproduced by AI easily.

[/quote]

Originally Posted by warrioraks

What would be the basis of this statement?

Repeating my point from the previous post - hallucinations (and therefore creating new stuff) is a feature of these GenAI models. People are using it to create new songs, deepfakes, images of Trump and Harris on the wedding aisle, and what not. Although I don't condone a lot of these things, the fact is machines are able to create stuff which was not part of their training dataset.

Creating new things is not what people think it is like. Giving an example in terms of automobiles again (since its teambhp forum!), give the world data of the era before the EVs came out, to AI and ask it to innovate the automobile industry. It will give you 1000s of engineering techniques (which is already out there on the internet) to improve ICE. Not one AI model will suggest moving to EV cause it simply can't come up with this new thing out of the blue. That's where humans are required. Believe it or not, the example given looks like a broader and bigger picture. But these similar things apply largely to regular day to day software engineering jobs as well wherein creative thinking is the key and no AI can solve it no matter what (in the similar lines of the example stated).