In my opinion, many of those people don't actually understand what AI is and what it is actually useful for.. They have not yet understood that "intelligence" is not actually what is provided.

Instead they have in front of them a very fast search engine that quickly collates, filters and sorts results that it has gleaned from the web and possibly one or two other resources... But, it cannot invent that which it does not know about, or make useful abstractions, nor can it provide answer to questions that are badly formulated.. And the big problem is with the data that it does have, if that data is skewed then you will never get anything but a skewed reply and the user does not know this

One of my non-it collègues has managed to build a simple website using AI results, there is nothing fantastic here, but he is unable to understand why it is so difficult to make minor adjustments, he has no formal training or experience on how or which changes to the HTML, Javascript or CSS. So he spends hours trying to get the AI engine to change the code for him but because he does not understand the code behind the site and therefore does not know how to ask the engine how to make the suitable ajustements.

He is using the a specialized tool for a job that he does not know how to perform. It's like giving a calligraphy brush to a four year old and expecting them to be able to write Japanese poetry.

AI is a good tool when used in the appropriate hands, but almost useless when used by those with little or no knowledge. Some of them would simply be better of just using Google.

For what? Lovely big post explaining why it's shit, let's have some examples of it being a "good tool".

If I have to check what it's produced then it's not a "good tool", it's an "idiot assistant".

"For what? Lovely big post explaining why it's shit, let's have some examples of it being a "good tool"."

As an exemple I could cite the following .

I asked it to produce code for a full screen Alarm Clock application, which allowed me to adjust the brightness by touching the lower/upper part of the screen, to automatically dim the brightness between 21:00 and 06:00 and to work in both Vertical and Horizontal modes, I use it as bed side clock on an old Android phone.

I am a coder so I understand the code that it produced, all I had to do was modify the Font for the clock face and Bobs your Auntie. It took me around 15 minutes to have a working application. I certainly couldn't have coded that it 15 minutes, I don't write Android code for a living, it would have taken me longer than that just to look up example of working code, contraints by constraint ..

It is a good example of where using AI to produce Boiler Plate code, which requires only minimal modification in order to produce a working application..

Obviously this is not what you would want to do for critical applications but for the boring stuff it actually works quite well. But again, the tool has to be used by those that already know/have an excellent grasp of what they are doing.

I am not an advocate for AI, for the majority it is an over-hyped fast search engine with some extra triage/filtering.

Just the other day I use ChatGPT-4o to write a simple Python tool. The prompt was (roughly):

> "Using Python write a tool that generates a Windows toast notification when a new serial port is connected to my system. The tool should run in the background and not require any user interaction once started."

I have a lot of serial ports on my system and going into Device Manager was a chore to find which one I had just connected. (Lots of FTDI cables, for anyone familiar with embedded systems.)

It created that tool in 30 seconds and it worked perfectly first time.

Could I have written the tool myself - absolutely. And I expect such tools are available online - but IT policies would mean I would need to get approval to install anything. So this saved me time. Instead of spending 10 minutes writing it, looking up how to generate 'toasts' on Windows, how to use pyserial to find the newest serial ports connected, I used a tool which has that information in its training data set and can generate the answer in seconds.

Would I use it for any serious production code? No. Would I use it for anything where I couldn't give the code a good look over and make sure it isn't doing anything bad? No. But it's absolutely brilliant for small, well-defined tasks.

Wait. IT policy requires that you get things approved before installing, but you don't think that applies to AI generated code?

No, because I have Python installed already and I can review the AI generated code to see if it's not a problem. 30 lines of code, not exactly a huge amount to look through.

The issue comes about when getting permission to install an EXE, I can get it, but it takes time. So I have a lot of tools written in Python for little problems here and there.

(Blame Cyber Essentials rules. Not my rules!)

Seriously, people would let AI write their emails for them? If the Email is so unimportant that you feel you can palm it off to AI, why are you bothering to send it in the first place? if it's a standard boilerplate response. "Thank you for your application, yadda, yadda, yadda". You can create a Template. Hell do it properly and you dont even need to have to modify it beyond adding the email address.

Transcribing Meetings? Who on Earth does this? And if you actually had to do this, would you really trust the hallucinations of an AI to actually accurately reflect what was covered in the meeting?

Managing files? If you are actually at the level of properly managing files (not just deleting the rubbish bin occasionally), then you would be far better off creating a script to do the managing for you. It's not going to suddenly go off and start "managing" other files that werent in the original remit, because the AI had a hallucination and decided that Root was a dirty word and therefore all files in Root had to be deleted...

I've yet to hear of an ACTUAL daily usecase for AI for the normal consumer or even the normal worker. It's just another buzzword used to try and sell new PCs. It needs to go away... Fast....

And this happens:

https://www.theregister.com/2024/11/21/new_york_times_lawyers_openai/

"The bot ate my homework."

I don't understand.

Just because the PC has an AI label stuck on it does that mean that the PC is incapable of functioning as a standard non-AI machine? Or is it the case that managers have told their staff "We have given you this new AI PC and you must use it."? Without training.

Or is it the case that the staff have been given this new toy and being curious have started playing around seeing what the new wonder weapon can do? Again without training.

It seems to me that management have fallen for the promise of increased productivity but have forgotten that a new method of working does require a modicum of training before they reap any rewards.

Sure, you can ignore the AI parts, but it's still a part of the silicon die that could be better used for something else. And that's if you have an OS that lets you ignore it.

I suspect that Windows will be full of unwanted AI "goodness" getting in the way.. Remember clippy?

Presumably the AI-written email will get an AI-written response, which gets a further AI-written response which gets ... ad infinitum.

Writing emails takes me no time at all compared to deciding what to put in them.

Transcribing Meetings? Who on Earth does this?

People who need a record of what was said and by whom for the avoidance of subsequent doubt or denial. Though in those cases you'd probably need each of the participants to review the transcript and accept its accuracy, so it's not clear AI would save much time in those circumstances.

"Just press record"!

In terms of transcription, I'm unsure whether the Teams service is deemed """"AI"""" but having had to interact with other regions has been outstanding. I've been able to (with local HR) recruit IT positions in other countries where it's allowed the candidates in a stressful situation to not have to add the burden of a 2nd language. I've followed the technical and visual ques with HR providing a localised view on their "feel and fit". The candidates are going to be based in that region and speaking in that language anyway, just need some English to interact with the UK team when not under pressure.

I'm not jumping on the "AI hype train" but the live transcription (if a little crappy at times) gets the job done really well.

I seem to spend half my time on the work computer arguing with it and to be honest I'm getting a bit fed up of searching for how to turn things off. I'm pretty sure that I don't need my writing scanning for robot-perceived faults: to quote Gilbert and Sullivan's 'Iolanthe': "I can spell all the words that I use, and my grammar’s as good as my neighbours’."

The robot complaining that the colour contrast of my email - y'know, standard black on white - might be a bit low for some readers is a final insult.

[Unsurprisingly, searching directly for that quotation to confirm it didn't find a hit; mostly I got online grammar checkers, or (with quote marks) nothing at all. I had to search for a libretto for Iolanthe and then search that...]

Been a long while since search engines actually search for the words that you've typed in, in order - they've been awful for finding lyrics in songs for a long time. To an extent, finding synonyms and "natural searching" is useful, but it would be very handy to turn it off sometimes.

"Text colour" is one of the things that LaTeX is quite concerned about, in the "traditional booksetting" kind of way. Rather than RGB colour, it means doing the hyphenation and layout in such a way that there's no big white gaps between words, or running gaps between words matching up over several lines, or squishing words to fit onto a line in such a way that there's a "dark line". That distracts the eye when scanning over the text, makes it difficult to read. Subtle if you don't know to look for it, but it's among the ways that a really good typesetter like LaTeX distinguishes itself from a really shit one, like eg. Word.

Dunno why you'd get that complaint when writing an email, though - you write the words, computer does the formatting.

If you want a search engine to find matches for all your words in order then you are after an exact phrase. If you format your search terms correctly then your search will be extremely effective.

If find they get song lyrics very easily.

Neil Barney’s is correct about the Iothanthe lyric/quote (try it, Google only found the quote when I prefixed the quote with Iolanthe.

However, the lyric “ Steal a lot and they make you king” Google does correctly find without additional prompting.

Been a long while since search engines actually search for the words that you've typed in, in order

Put them inside quote marks. Job done.

That works for the Iolanthe quote.

Text colour RGB.

I remember reading, when still of an impressionable age, a page of text sporting "meaningfully" coloured words in Michael Palin's [1]Bert Fegg's Nasty book for boys and girls ☆ which, I think, featured the titular Dr Fegg and his axe but I vividly recall red as in blood was greatly in evidence.*

While not being at all squeamish in the matters of blood and gore, that experience induced a lifelong aversion to varicoloured text. Chromokeimenophobia ?

* along with the trite examples like sky, sun,...

☆ Methuen 1974.

[1] https://en.m.wikipedia.org/wiki/Bert_Fegg's_Nasty_Book_for_Boys_and_Girls

So.... when are you changing your name to 'Dave'? (2001 reference for those not old enough to have seen it in the cinema first time around)

I have spent the last 50+ years working on computers and stuff, and I will refuse to speak to my frigging machine even though in a past life, I actually worked on some voice recognition software (oh the shame of it)

The like of MS want to dictate how we should work so desperately that they make it harder and harder to block all their crap and turn it off. I had to spend a whole day last week doing just that for a friend who had a new laptop. No, I don't want to search the entire internet for everything I want to do. Dumbmotherfsckers.

All that crap is just one of the many reasons why I dumped MS for personal use in 2009. My go to machine is currently an M1 MacBook Pro. I find MacOS far less intrusive that anything to come out of Redmond since Windows 7 (once you tamed it)

Yeah, I too dumped Redmond about fifteen or twenty years ago - so long back that I don't recall, though some software I wrote about twenty years ago still follows me around and I have to field enquiries explaining that I don't do Windows any more.

Sadly, work still feels the need for Windows.

>” even though in a past life, I actually worked on some voice recognition software (oh the shame of it)”

I was wondering whether part of the problem with AI is the same as the problem with the “open mic” of say anything continuous speech recognition applications.

《Unsurprisingly, searching directly for that quotation to confirm it didn't find a hit; mostly I got online grammar checkers, or (with quote marks) nothing at all. I had to search for a libretto for Iolanthe and then search that...》

DuckDuck quacked on

Gilbert and Sullivan "I can spell all the words that I use, and my grammar's as good as my neighbours'."

on the third entry after two adverts (grammar training, then the usual ebay bait [& switch] :)

[1]https://gsarchive.net/iolanthe/web_op/iol08.html

I am surprised that Bing wasn't too shabby at this, particularly given it's a Microsoft creation.

Looking at the libretto, some of the versification sails pretty close to doggerel. Tabors/Neighbours, Hearty/Party - end rhymes to die for, or do I mean lethal? If I were Gilbert I mightn't set my hopes on Poet Laureate (actually he wouldn't have been the worst.)

[1] https://gsarchive.net/iolanthe/web_op/iol08.html

I use DDG. My first attempt just had the quote, no quotes. That used to be a good way to get a lyric if you'd forgotten a word or two... nada.

Adding quotes provided 'no results match...'

Removing the quotes and adding Gilbert and Sullivan _now_ drops me straight into the GSarchive page you got - but the first time, just repeated the grammar checker adverts.

Wasn't it nice when search engines were deterministic?

I'd love to see a study that has a look there, and it's not just bad UI design.

There's also the confusing mess of OneDrive vs SharePoint, the active encouraging of shadow IT (certainly with how they currently are flogging Copilot) which means that people with no actual competence can now take their mischief beyond Excel and make decision tools that run on near impossible to audit models, I can go on. And yes, that's before mentioning the ever present waste of time by eternal patching. You know that a version of Windows is approaching something of stability (no, I said approaching ) when Microsoft starts preparing a new version.

If MS were as competent at writing code as they were at lobbying, marketing and manipulation their products could actually be useful, but code dev stops at 'nearly good enough'after which it is rammed down the throat of its victims users who no longer have any choice but to be beta testers and thus be on the receiving end of a style of quality control that even 90's era Lada would have been ashamed of.

And yes, I didn't have my coffee yet. How did you know?

Y'know, who would be the types that got max-spec brand-new PCs already?

We now have a worthy successor to "Powerpoint makes people stupid"

Talking to some bank / isp / mobile company chat bot: “I want to talk to a human. Don’t tell me I can find answers on your web site because I can’t so I’m talking to you instead”.

Talking to a misbehaving computer: “Right, you asked for it. I’m going to give you a damn good thrashing.” *fetches very big stick*

Yet another load of hype for something most of us have little use for, yet they invested billions in it and need to recoup their costs, so we get even more hype, and the bubble grows and grows until POP! I see uses for LLMs in image analysis and data mining but whatever they do needs to be checked by other means because they are untrustworthy, and yes, so are humans. I REALLY wish that the marketing idiots would stop using the term AI because it isn't and never has been Intelligent. Perhaps it actually means Artificially Incompetent? I am currently in the market to replace my home/gaming PC and I actively avoid any that even mention AI because they have obviously bought into the hype and I will be paying over the odds for something I don't want. I already waste far too much time CORRECTING my PCs when they do something for me that I DO NOT WANT them to do, just because some muppet decided it would be a good idea that everyone on the world is forced to work exactly like they do. The LAST thing I want is some overblown Clippy trying to "assist" me.

"Our role as technology leaders is to support this transition to AI-assisted living.... "

I'm not old enough that I need assisted living thank you very much.

Now kindly fuck off.

Aka : please help us sell this new generation of bullshit.

"44 percent of respondents believe AI PCs are a gimmick or futuristic""

Really,,? I thought it'd have been far higher.

" "Many AI users spend a long time identifying how best to communicate with AI tools to get the desired answers or response "

This has for ages been the bane of those trying to get meaningful results from search engines. The evidence is out there -- just nobody in the technocracy (or more realistically, the tech plutocracy) has bothered to look.

Generative AI produces bullshit: plausible-looking output that may or may not bear any resemblance to reality.

Luckily people are quickly realising this, and the bubble hopefully won't inflate too far before it bursts.

AI, as a tool for spotting patterns contained in mass datasets, _does_ have powerful uses.

The "I" in AI cannot stand for "Intelligence" as intelligence isn't involved anywhere. It should be called "AL" - Artificial Learning. Or "APM" - Artificial Pattern Matching.

Yeah, I've seen [1]that film .

[1] https://en.wikipedia.org/wiki/Demon_Seed