[asingh1977]

Originally Posted by SmartCat

Do IT companies allow employees to use generative AI for work? Aren't they worried about secrecy of their client's software/code?

Good question!

Depends on the paranoia level. Product companies, I know are quite restrictive and are not moving into an open-usage field. Specially if the product company is handling PII (Personally Identifiable Information), it is a general sense: that this information maybe posted to a GenAI model and stored, or tokenized / embedded. (i.e. start picking up prompts and repurposed for RAG (retrieval augmented generative AI) or learning models). All LLMs, do not exhibit this behavior, but if the model is not running on premise (read is online), at times organizations fear their usage. So the restrictions.

On the other hand a lot of service / consultancy based firms allow usage of GenAI products, they see an immediate boost in employee throughput and a lot of mundane tasks are either quicker or automated. And there is always the chase to get a full blown product out to market based off GenAI methodology. This is where most companies are struggling. They are able to develop products, but adoption of foundational GenAI products is still nascent, vs. the big 5-6 which have released transformer models based on some sort of minimal billing/membership. They have seen success.

Many companies are now realizing that the period of harnessing a large LLM model, trying to put a wrapper around it, augmenting with APIs, taking this to sell: is basically over. Notching up cloud provider infra. costs is also an issue with end-users, who have started to question the RoI of: proof of concept, build, deploy, and hosting costs.

Times are exciting no doubt. There is a lack of quality engineering around GenAI / LLMs (read people), so people who are ready to learn and correct course early, are the ones who can benefit from all this!

[shibulijack]

MOD NOTE : Please do NOT post any responses from AI / ChatGPT. We take pride in our original thoughts, opinions & content.

Thank you

[SpeedBird]

The one thing "AI" can never do, is create new things. AI by definition works on feeding on existing things and presenting it either as the same or as a blend of many results, making it look more useful or appealing. The job roles such as L1, L2 support, middleman roles in corporates etc, who are basically echoing information by searching from somewhere, are the ones which will be affected. To those non-tech people who are excited to finding this "technical co-founder", they will soon understand that they are just redoing things and are essentially not creating anything new. This is fine as long as the expected result is, lets say a website front end or something which doesn't require uniqueness and much of backend processing. But with thousands of these new startup CEOs doing the same with the same AI and same data points, the results are bound to be same. Hence the majority are more likely to be closed down, in the same way it happened during dotcom bubble. This is what we call as the "AI bubble" and by the time the new grads learn about proper AI backend, the trend of AI itself would be similar to the trend of e-commerce and networking today (there will be one or two monopoly companies and all other small start-ups would have been closed).

The only safe software engineering jobs which do not come into the above mentioned umbrella are the core software jobs, which requires knowledge of working of the system at core level, wherein most of the code is undocumented, changes from one chip to another and are mostly proprietary. This is the proper software engineering job wherein google and stack overflow won't help you. This is the stream on computer engineering which will survive and keep surviving. (Ironically, this is the field which the "experts" are calling obsolete whereas this is the baseline on which the entire AI agenda is running on. One major bug in the foundation and no experts to fix it, the entire building collapses. New grads focus should be this very field). All the other dependant roles, such as support role, middleman role, including even the "gen AI" bubble role, will be obsolete within the next decade. History is the proof of this. Calculator didn't diminish the mathematicians, code generators didn't diminish the programmers, and even e-commerce didn't diminish the local vendors. Today, AI won't diminish the software engineers.

[rkv_2401]

Originally Posted by SmartCat

Do IT companies allow employees to use generative AI for work? Aren't they worried about secrecy of their client's software/code?

You don't need to provide the LLM access to your codebase to get useful answers. Although you can use IDE-integrated AI tools such as CoPilot for VSCode and the Cursor IDE, just writing a prompt without sharing specifics about your code (similar to how you would Google to find answers before) can tell you everything you need to know from a software development POV.

There are engineering blogs that show that being able to parse the entire codebase into a syntax tree (which the model has access to) enhances the quality of responses of the model. But I personally feel the IDE integration of AI is just laziness. There are occasions when I need AI at work, but I don't need to sell my company's codebase (soul) to allow a model that is accurate 60% of the time to instead be accurate 65% of the time. Of course, if your codebase is a generic web app, maybe you're more time sensitive and the secrets of your codebase (if any) aren't vital to the survival of your company, you'd be inclined to spend the $20 a month on Cursor IDE and get the development over with ASAP for an earlier launch.

I'm not sure about the legalities of using LLM "generated" (i.e. regurgitated) code in a commercial codebase or things like that, lazy employees feeding the models PII or proprietary algorithms/data, etc. Which is probably why there are blanket rules covering the use of AI in larger organizations.

[somersault]

Originally Posted by bordeaux

Only a few points I have on this:

• Blockchain was going to solve every problem on earth.
• Bitcoin was going to solve every problem on earth
• We will only need to own NFTs. No real life work required.

Add to this the following,

• Internet of Things (IOT) will change the way we operate.
• Crypto will be the only currency going forward.
• ML/AI/Deep Tech is some state of the art creation from out of the world.

If you use any of the buzzwords, your startup is going to be funded soon.

[SteadySteer]

Originally Posted by SpeedBird

The one thing "AI" can never do, is create new things.

Well said! AI excels at generating variations of existing work rather than creating entirely new things. It’s likely to take over many routine jobs that have persisted despite extensive automation, but this shift will also open up new opportunities. It’s not the end of the world—just a change in how we work.

[ValarMorghulis]

Originally Posted by SpeedBird

The one thing "AI" can never do, is create new things...

Originally Posted by SteadySteer

Well said! AI excels at generating variations of existing work rather than creating entirely new things...

May I add a few words?

One thing "AI" can never do, is create new things...for now.

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

[Janus]

Originally Posted by AMG Power

One of the drawbacks of human intellect is its inability to come to terms with the fact that there is a potential for an intelligence superior to its own and that there are limits to human intellect.

The inability to accept this fact is in itself a limitation of human intelligence.

Rational thinking, scientific proof, one’s own experience are the usual arguments in support.

Most of the statements on AI are based on its current state of evolution and what humans can conceptualise as to its potential future.

Unlike traditional AI, which relies on predefined algorithms and human crafted updates, self-evolving AI harnesses the power of evolutionary algorithms (and the usual Machine learning & deep learning) to enhance its own performance over time.

This evolution is not just about handling more data or solving more complex problems; it’s about AI’s ability to learn from new situations, environments, and challenges (as humans do) thereby becoming increasingly similar to human thinking and eventually moving ahead of humans as it evolves.

Some would say that this is conjecture.

I hope it is.

Perfectly said. If you ask about AI to anyone now, they will reply based on their 40 years experience or current available technology and its linear growth of what it can do in the next few years. It's just an if-else and it cannot create on its own or at-most it can takeaway L1 jobs.

No one thinks about the disruptor.

How many of us have worked to make their own job obsolete, no one wants to let their work go because they are scared at being made redundant, lose salaries, no money to pay loans or feed their kids.

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

For example, if we build a brand new bank from scratch with modern tech and migrate/merge all our 12 PSB's into this new bank, we will have one super bank providing all the facilities and in the process make 1000s of jobs redundant.
People focus on the 1000s of jobs lost but not on the new economy created with one super efficient modern bank and its advantages.

Similarly there will be someone who thinks like this and once the next generation Artificial General Intelligence takes over, our IT industry will be caught pants down.

[DudeWithaFiat]

Originally Posted by somersault

Add to this the following,[list][*]Internet of Things (IOT) will change the way we operate.
......

Isn't IoT actually changing lots of things? Connected cars, smart watches, smart locks, smart TV....

[mayukh42]

Originally Posted by Jeroen

I never understood this software engineer hype ... If you can afford to shed that many people you are providing a pretty common type of service I would think?
...
It used to take a master from a technical university and at least 5 years of supervised experience before you would be allowed to touch code
...
A lot of what is called programming and coding today, is no more then relative simple configuration.

The "engineering" (i.e. solving problems) part of software industry is very similar to other engineering industries. This space needs domain knowledge, and years of expertise to obtain a solution. Most "product" companies that do this, can easily be identified by the higher age of employees. Coding is a very small part of the effort, and even if part of it is replaced by AI, it will only make the employees focus on smarter things.

The other part is what has created this hype - "shops" where known solutions are wrapped into new packages and sold as fresh products. Built on proven paradigms, then themed as per the target industry or current trends, Mostly a "Cloud" or "e-" prefix to the product name is all that is required to create the hype (nowadays, "AI" suffix).

Such products backed by uber-rich venture capitalists whose only motivation is to snatch the customer base of their "friends", have created a bad name for the industry with sky-high salaries for even freshers which were never sustainable in the long run. These are like "hit-and-run" businesses, attract customers, call it "disruption", get more series of fundings, after which one fine day when predicted growth does not match, either sell it off or close shop.

It is not that the industry has not at all gained from them. Some of these companies have also contributed to the technology, created frameworks that others have adapted, so engineers working there are not always different from those in the first category. But they are far and few, and most are just here to capitalize on the next "buzzword", which is ever growing list -

• dotcom,
• internet (i-*/ e-*),
• big data,
• blockchain,
• and now AI.

Each of them created a bubble that was corrected over time.

Originally Posted by Jeroen

There are plenty of people who choose their career based on very different parameters.

Same in the software industry too; many are here because we like it, not because of hype. The "satisfaction" when the code compiles and runs as expected is unbeaten since the days of primary school. Such people can be found across the spectrum, since career interests are not always explained by the company one works for.

Software industry can boast of the largest group of freelancers who have made consumer/ enterprise grade software like Linux possible, without being on the payroll of a full time job. No other industry has this much "community" output outside of standard paying jobs.

Originally Posted by Janus

Ask about AI to anyone now, they will reply based on their 40 years experience or current available technology and its linear growth

Slightly OT: AI is a subset of machine learning, It is not new. Only its applications are more popular now because most of the AI models available, made by the "cloud" companies (Google, Meta, OpenAI i.e. Microsoft) who collected data from devices over a decade could also have the hardware now to train the models efficiently, unlike in the 90s.

Similar things that existed in the industry but only recently found traction are distributed systems, multi-tenancy, virtual machines. Hardware, network bandwidth becoming cheap are what promoted them to the forefront.

Originally Posted by Janus

How many of us have worked to make their own job obsolete, ... Only a handful, in my 15 years of experience, are willing to migrate away, simplify, minimize and move on to the next thing.

Software engineers are in fact expected to keep up with technology and adapt with the next. The industry is not even hundred years old, unlike other branches, and things are still moving at a rapid pace. For example, shared libraries are a thing of the past, nowadays "fat" executables having all dependencies bundled, or "containerized" apps like those on Apple devices, with multiple redundancies but more resilience, are the norm.

Software industry is actually the one where technology moves the fastest, in other industries the engineers don't have to pivot that frequently. A typical technology "generation" is 5-7 years at most.

Originally Posted by Janus

make 1000s of jobs redundant, ... new economy created with one super efficient modern bank and its advantages.

Computers posed similar concerns in the 90s, but the only thing happened is that the humans moved to more intelligent jobs, or could do more in the same time. Most software engineers will adapt as they always have in this industry more than others.

Redundancy will be trimmed of course, and hopefully it will nullify the hype which adversely affects the industry, as most of the executive leadership live quarter to quarter and decide based on impact to stock prices. The industry needs to be more mature and think longer term.

[somersault]

Originally Posted by DudeWithaFiat

Isn't IoT actually changing lots of things? Connected cars, smart watches, smart locks, smart TV....

Lot of things will always be new to our way of working. The question is, whether it is a large scale disruption as marketed by those technology evangelists or just an innovation.

IOT was the buzz word some 3 years back in the startup industry. But not anymore. The impact is limited.

[hsaxena]

Originally Posted by SmartCat

Do IT companies allow employees to use generative AI for work? Aren't they worried about secrecy of their client's software/code?

I work for a GCC and my parent company has an in-house tailored Enterprise GenAI specifically to be used by the employees across functional units. Since my org is in Healthcare Insurance space, the policies around PII, PHI are pretty water tight. While we do have access to this platform any other code writing LLMs like copilot, Claude etc. are all banned.

As a developer, the improvement is noticeable, but not life changing. Most of the work on enterprise code revolves around working on large codebases, migrating to new tools, rewriting functionalities for optimisation. The platform has essentially replaced Google. I converse with it like a buddy to help get me answers (something like what could cause the node to fail in xyz setup etc.), which are not precise but can help in troubleshooting which would have otherwise been an oversight.

For code generation, I like boilerplate code that is easily available so I don't have to write 100s of lines easy functionalities, but implementing core logic eg. optimising 1000s of lines of complex SQLs, or performance testing, code profiling etc. are still done manually yet (albeit with some help from the platform). Yet because I see almost all software engineering touchpoints, being automated or at least supported by tailored OSS LLMs that will be available a dime a dozen running locally.

Software developers of the future (10-15 years tops) will be coding directly using a language closer to English than to machine. Analogous to how we don't punch cards or write chip specific machine instructions but build platforms using a framework like React, Tensorflow etc.

[am1m]

Originally Posted by hsaxena

I work for a GCC and my parent company has an in-house tailored Enterprise GenAI specifically to be used by the employees across functional units

...any other code writing LLMs like copilot, Claude etc. are all banned.

As a developer, the improvement is noticeable, but not life changing.

What is your take on the quality of output of the company's in-house AI versus other mainstream LLMs like Copilot, etc?

I worked for a company that used an in-house developed AI too (though the base LLM model was still a commercially available one) mainly because of security concerns. But frankly, the output sucked when compared to the other commercially available options. Understandable, LLM development was not their forte. This was more akin to a good college project, looked good on the resumes of those who developed it, but not for much else!

I was reminded of the time, a long time ago, when I used to work for a company that insisted on developing their own search capability instead of "exposing their content to Google", again because of privacy and security concerns. As you can imagine, that idea has aged so well!

I guess the same will happen with commercially available AI models over time.

[hsaxena]

Originally Posted by am1m

What is your take on the quality of output of the company's in-house AI versus other mainstream LLMs like Copilot, etc?

The output is as good as GPT3 if not better. Better because it's RAG'd to our massive confluence database among other knowledge sources. This is a proper Enterprise solution, costing at least 7 figures to the Enterprise if not more.

Originally Posted by am1m

I was reminded of the time, a long time ago, when I used to work for a company that insisted on developing their own search capability instead of "exposing their content to Google", again because of privacy and security concerns. As you can imagine, that idea has aged so well!

This is exactly how I see it. The consequence of this is the expectations of productivity, for any role in general, is many fold more. You don't know the exact definition of a KPI, GenAI will give it to you, you don't have to scour through massive Knowledge libraries. You don't know what changes are needed for a tool upgrading a version, GenAI will give it to you. All of this with an obvious downside, if it starts to hallucinate, you will, in all possibility, will mess up one thing or the other.

[JediKnight]

Coding has been misunderstood as computer science/software engineering. It is like calling the service center technicians as engineers as many people do.

Real software engineering is very complex task and requires very high mathematical/abstract reasoning capability. There will always be premium for real software engineering. And it is different from ability to code in a high level language which is just a small part.

For people who are just "assembling" programs by calling high level APIs, the so called AI can be a serious competitor.