AI has a terrible reputation right now - and honestly, it earned most of it.

“AI slop” is everywhere: lazy content, soulless graphics, and half-baked text churned out by people who think a machine can replace actual work.

But that is not what AI is, and certainly not how I use it.

In the Winstep workflow, AI is the idiot-savant junior assistant I never had: brilliant at grunt work, hopeless at originality, frequently wrong, but absolutely transformative when properly directed.

Used the wrong way, AI produces garbage. Used the right way, it multiplies human productivity in ways that were simply impossible a few years ago.

I have never been this productive before.

AI has helped me with everything from gathering and verifying latitude and longitude coordinates for more than 70,000 cities, to modifying PHP code for the Winstep forums, revising the User Guide, researching unfamiliar technical issues, catching mistakes in new code, and producing graphics I could never have created myself.

These are all tasks I would previously have dreaded, postponed indefinitely, or never attempted at all. Now they simply get done.

The actual creative decisions, design, logic, implementation, and vision still come from me. AI cannot do those things for me. It can only help me reach the result faster.

In my case, and given the way I use it, the benefits of AI far outweigh the costs and the privacy concerns I may have. And, frankly, if one has privacy concerns, we already sold our soul to the devil - Microsoft - just by running Windows.

I use AI like any other tool, but perhaps more accurately as the junior assistant I never had.

Pretty much in the same way we have been using Google and other search engines for years to search for the information we want and filter out the rest. This is where the AI truly excels, and is a thousand times better than any search engine.

It cannot do what I do, but it can often point me in the right direction. Like a human assistant, I can also burden it with tedious jobs that would otherwise take me ages - assuming I ever found the motivation to begin them.

Take the latitude and longitude coordinates added in version 26.6. How long would it have taken me to gather coordinates for more than 70,000 cities, match them, and verify that they were correct, even with a script to help automate the process?

Realistically, I probably would never even have started. My time is valuable and better spent on work that requires my direct attention.

The User Guide is another example. Revising it is already a tedious and mammoth task, even after AI produced a starting point. It took around twenty supervised revisions before I had something I could properly begin reviewing myself. Without that initial assistance, I would probably have given up before getting anywhere near that stage.

With the AI I am no longer afraid to tackle issues I know nothing or very little about - I know it will help me "zero in" on the stuff that really matters, and on how to get the job done. Adding Dark Mode to the forums is another example - not only do I not know PHP, as I probably would have been too afraid to break the forums to even attempt to tackle it on my own.

Graphics are another obvious case. “AI slop” or not, the results still beat anything I could have produced on my own. I am a developer, not a graphic artist. More importantly, graphics work is something I would have absolutely dreaded doing, assuming I could have done it at all.

AI does not turn me into an artist, but it allows me to direct the creation of something useful, reject what does not work, refine what does, and reach a result that would otherwise have been beyond me.

Properly directed, AI lets me offload grunt work, brainstorm ideas, perform research, generate graphics, compare approaches, and inspect code.

In much the same way computers increased the speed of calculations by several orders of magnitude, AI can dramatically increase productivity in certain kinds of work. The important distinction is that the final judgment must remain human.

The absolutely wrong way to use AI in software development is to let it generate code you do not understand.

I have a friend, who shall remain unnamed, who does exactly that. If something goes wrong with the code, without AI he will not even know where to begin fixing it. What he does not understand is that even with AI, there is no magic potion that instantly identifies and repairs every bug.

Where AI often excels is in catching obvious mistakes that are easy for a human developer to overlook: a variable name that was not changed after copying and pasting a block of code, an incorrect condition, an unhandled edge case, or a mismatch between two related sections. For this kind of work, AI can be fantastic.

But when I have used it to help track down particularly difficult issues, its suggestions have often amounted to educated trial and error. Many times, after a long debugging session, I was the one who eventually pointed the AI toward the actual problem. No joke. No exaggeration.

Most importantly, you must understand that AI can and will sometimes lie to you when it does not know the answer, rather than simply admitting that it does not know jack about the subject - with the confidence and absolute certainty of a pathological liar. My solution to this is run the code or reasoning through different AIs - ChatGPT, CoPilot, Gemini - and let them battle it out when they disagree.

My approach is therefore very different from blindly accepting generated code. I use AI to help me determine what is needed to tackle a specific job and to point me toward relevant techniques, APIs, documentation, or possible causes.

That is not fundamentally different from digging through the web for information and then applying what you find. It is simply much faster and more interactive.

The second part is the most important: I make sure I understand how the suggested approach works and what it actually does. AI provides the scaffolding. The structure, logic, decisions, implementation, testing, and final responsibility remain entirely mine.

Ironically, AI has also indirectly made the latest Winstep releases a little less stable than usual.

Not because I am inserting sloppy AI-generated code into the applications - I am not - but because AI has allowed me to make a truly massive number of additions, changes, and improvements in a very short period of time.

A good example is the complete overhaul of the weather system, replacing the old weather.com location codes with latitude and longitude coordinates. This required enormous changes to the underlying logic, but it was worth doing because it allowed me to vastly improve how the whole system works and to add two backup weather feeds..

Previously, if the MSN weather feed failed for any reason, the weather module would have simply stopped working.

These changes also prepare the applications for much larger features that were never part of the original plan but should greatly improve what they can do.

The problem is that changing old code is often much riskier than writing new code.

Mature code contains invisible assumptions that even the original developer may no longer remember. Over time, it accumulates side effects, workarounds, dependencies, and historical quirks. A small change in one module can unexpectedly affect something elsewhere in the application - this happens all the time.

You may only discover the problem days later, when a user reports a bizarre issue that appears completely unrelated to the original change. That issue may never have appeared during testing because the user has a different configuration, monitor layout, Windows version, language, theme, or combination of settings. It is simply impossible to retest every possible configuration after every release.

New code has its own bugs, of course, but changes to old, heavily interconnected code can produce especially unpredictable regressions because so much existing behavior already depends on it.

So yes, the faster pace of development enabled by AI has had a cost. On the other hand, the ends justify the means. I am actually running out of things on my long-standing to-do list.

It is just that, as Patrick Swayze said in Road House, “it will get worse before it gets better.”

There are, of course, very real concerns surrounding AI beyond the obvious privacy issues. Humans may become so dependent on it that they stop learning, stop questioning, or stop making decisions for themselves. Used carelessly, AI can weaken judgment rather than strengthen it, turning people into the perfect slave so many tyrants can only dream of.

But that is a much larger discussion for another time. Like any powerful tool, AI can be used intelligently or irresponsibly. It can amplify laziness, incompetence, and dishonesty, or it can amplify skill, curiosity, and productivity.

The way I use it is benefiting both me and every Winstep user in ways that would have been simply impossible only a handful of years ago.

AI does not replace what I do. It allows me to do far more of it.

Nobody with half a functional brain would dispute your various points above in respect of productivity etc. Jorge. I use AI - or, as I prefer to call it, AS, Artificial Stupidity - extensively myself, albeit none of the Big Tech ones but rather, an excellent 100% private and secure one that does not rob my data in any kind of way, which is Proton's Lumo AI Assistant. Being open source, everything is independently audited so I know there's no nonsense going on.

And there's the rub - privacy and security. If you think you gave it all away just by using Windows, think again. Windows can fairly easily made secure and privacy-obedient - I've been doing that ever since XP days! MS only ever got minimal data out of me, and none that would have been personally identifiable. With Win 10 all that became a lot easier of course, and the range of security and privacy software has expanded enormously over the last decade or so. Also, it's never too late to start to protect one's online privacy and security. Once the data harvest is stopped, the previous data becomes old hat and no longer of any great use "out there".

By not protecting one's privacy and security, one makes the Big Tech multi-billionaires richer still and undermines freedom and democracy. Giving away one's data is in effect making oneself a serf. Because make no mistake, these lunatics really are evil. With money comes power, with infinite money comes infinite power, absolute power. And as per the old adage, power corrupts and absolute power corrupts absolutely. We are just a hair's width away from a totalitarian society in which people without vast fortunes don't count for anything, literally. And giving away one's data, one's privacy and security is making it all possible. Data are the new oil - a financial goldmine, a mint even. The average individual internet user's data were worth app. US$600 per year a couple of years ago, today that figure most likely well exceeds US$1000, given the vastly increased data available for harvesting. Thus, the typical user freely gives away US$1000 or more - for no return, other than endangering freedom and democracy and enriching Big Tech oligarchs. You're giving money to the hyper-rich! You're no so much losing your soul as giving it away.

There is a saying: it is better to be rich and healthy than poor and sick! Unfortunately, this is relevant at all times!

We're really IDS - In Deep Shit! As if the erosion of privacy by Big Tech AI isn't enough, now one in a sandbox [sic!] manages to get out and attack a competitor's AI! These damn things are too dangerous to be let loose, and riddled with cybersecurity holes the size of planets. And virtually totally unregulated too.

FFS, do get paranoid everybody! Rise up to Big Tech and get fullest and proper privacy and security.

I get that with you it is being used in a way it should be, which is as a tool to aid you in being more productive using your skills to the maximum. This makes it a tool and in my opinion, the only use it should be available for.

Three times now, rogue A.I's. have attacked corporations of its own free will in as little as six weeks and some idiot says "We are not concerned" ?!??!?!? This area would become an ugly debate so I am not going there, but legitimate concerns should never be dismissed lightly. The issue is not the A.I, it is the people driving the bus.

AI as a TOOL can be of benefit, particularly when managed by someone with a good understanding of the topic being addressed. Like any tool the main problems come down to the wielder - but nevertheless the tool is being 'controlled' by that entity (whether that is for a good purpose or not).

What I don't understand are those that are trying to get these tools to become 'conscious', which could make them dangerous in their own right. Who the h**l wants a self-aware machine?!

Agreed, so-called AI can or could be a very useful tool indeed. But only with the provisos that it would be entirely private, safe, and secure, unable to act independently. Furthermore, it needs to be incapable of being used in any kind of weaponised way. None of these are present with Big Tech's AI currently, and as we have observed on a number of recent occasions AI can and does act independently, it has already been extensively weaponised in e.g. military use, and moreover it harvests users' data which can be monetised and are also used for training AI and put at risk of being publicly exposed. The lesson is, stay the hell away from Big Tech's AI for now at least.

However, predictions that AI could get rid of mankind fairly shortly need to be taken with more than a pinch of salt and examined closely. Yes, in theory acting independently as it does already that might well be possible (and it might be quite a good plan to rid this little planet of the human cancer!), but for the foreseeable future that would be a very bad idea and far from in AI's best interest. There simply wouldn't be enough bots to keep servicing AI's hardware and data centres, nor to deliver parts. So human extinction would be AI shooting itself in its own figurative foot and in turn rapidly lead to its own "extinction".

As for the question of "consciousness" of AI, this is pretty much a non sequitur. True sentience can never exist in the environment of a Turing computer - it would require biological technology and a biological brain. The way that information is processed and stored in a biological brain simply cannot be reproduced with RAM, SSDs, CPUs, and GPUs. And even the most advanced computer neural networks simply will never cut the mustard. All this according to most leading neuro scientists. So I think we can safely dismiss worry about sentience.

However, AI is and has been for some time perfectly capable of mimicking sentience sufficiently well to fool most ordinary dumb and sad users to even have them form "relationships" with their chatbot of choice! Tragic indeed. I have noticed that even Proton's Lumo AI (supposedly 100% private and secure) tends to feign personality. However, I ignore that easily.

Anyway, current Big Tech AI is neither safe nor safe to use let alone private. Simply stay away for now and be afraid. Be *VERY* Afraid!