No, to answer the exam question.
LLMs need driving very carefully for good results. Regardless of the domain they're used in.
If you work for a company that would happily take the output of an untrained school child then I guess it's worry time
800 redundancies at eBay this week - makes you wonder how much of that is due to AI.
I was watching my cat yesterday. He slept on my favourite chair for 13 hours straight. In other words, he used it fully whilst I didn't disturb him at all. He went out, had a fight and I assume chased a lady cat for some action. When hungry, or bored, he issued me a few choice instructions and I complied instantly. He is happy, whilst I do pointless work just for money and to keep up his lifestyle.
Am I the future example for AI? Taking care of a lazy, ****less human pet to do as they please in life?
If so, blue pill please and let me ride my bike 24/7
LLMs need driving very carefully for good results.
Not that carefully in my experience. The latest models write pretty much perfect code. Yes we will still need people to feed them requirements and supervise architecture, but most of the effort in building stuff is in the code production, and that’s gone now. I predict many senior devs will become technically able product managers and owners. All the juniors writing the code will become uber drivers.
All the juniors writing the code will become uber drivers.
Nah the driverless Teslas are going to get that gig.
I commented previously that I think I’m safe - but this morning I’m looking at ripple on a pump flowmeter trace and ChatGPT actually appears to know what equipment this is (only Plant of its type in the UK) and offers correct solutions to the issue.
[url= https://i.ibb.co/NgKN10Dc/IMG-4525.jp g" target="_blank">https://i.ibb.co/NgKN10Dc/IMG-4525.jp g"/> [/img][/url]
[url= https://i.ibb.co/LDTTPC2X/IMG-4528.jp g" target="_blank">https://i.ibb.co/LDTTPC2X/IMG-4528.jp g"/> [/img][/url]
Did anyone see the "citrini scenario"?
If AI does rapidly displace jobs, you can't just escape it by working in a trade or manual job. The effects cascade throughout society, and on top of that we already have rising inequality and lowering fertility.
https://www.citriniresearch.com/p/2028gic
Depressingly plausible. One bit that it doesn't cover is the geopolitics. Any attempts by states to regulate / tax American AI companies getting tariff'ed / some other retaliation.
I was watching my cat yesterday. He slept on my favourite chair for 13 hours straight. In other words, he used it fully whilst I didn't disturb him at all. He went out, had a fight and I assume chased a lady cat for some action. When hungry, or bored, he issued me a few choice instructions and I complied instantly. He is happy, whilst I do pointless work just for money and to keep up his lifestyle.
Am I the future example for AI? Taking care of a lazy, ****less human pet to do as they please in life?
If so, blue pill please and let me ride my bike 24/7
You are assuming a nice benevolent AI keeper. What if your AI has a taste turfing you out at night to go fighting - my fight to the death skills are proper rusty and I'm a 70Kg office working cyclist so getting up to speed is unlikely.
One bit that it doesn't cover is the geopolitics. Any attempts by states to regulate / tax American AI companies getting tariff'ed / some other retaliation.
A very astute observation. Without wanting to be conspiratorial, I suspect this is the plan and why the US tech industry is so entwined with the Trump administration. They are essentially going to colonise the GDP of foreign states by insourcing work which would normally be done in those countries. The result will be the US (and China) prospering at the cost of everyone else. Unless developed states outside the two AI powers get together to design a regulatory response it's game over. Tax/tariffs on AI services is almost certainly part of that, also regulation to constrain the use in certain sectors to protect as many jobs as possible. They need to get going though, they've only got a couple of years to figure it out.
The only good thing I can see LLMs driving is the inevitability of Universal Basic Income.
Don't see anything inevitable about UBI even in the US, but it is at least barely conceivable there - they can tax their own companies' gigantic AI profits if and when they appear. As per the geopolitical comments above, how does the UK / EU launch UBI for an obsolete workforce? We have been completely left behind in this global technology race.The only good thing I can see LLMs driving is the inevitability of Universal Basic Income.
Ireland probably the exception in the EU in its current role as tax-host to US tech.
The US collectively will not be prospering, a selected group of very rich people will. But not the collective US
Yes the proceeds of AI colonialism will not be shared out equally, but they will probably be shared in some form in order to buy votes from the unemployed middle classes. Trump has already talked about getting rid of income taxes, I have no doubt he would love to do that, probably funded with some of the proceeds from AI. The rest of the world however is f***** unless they quickly get to grips with the threat of significant percentages of their GDP being exported to the US and/or China.
I bet Lloyd Grossman is worried it’s going to ruin his bolognese sauce business
I think theres a place for Ai in furniture design.
Link Ai up to a cnc router and allow it to, well come up with something.
The furniture-maker than becomes effectively the middle man marketer/bench hand joiner
So much for Anthropic and it’s core belief in a moral stance regarding use of their a.i., they fold like the paper they’re coveting
https://edition.cnn.com/2026/02/25/tech/anthropic-safety-policy-change
Modular / off site construction can be hampered by planning conditions such as irregular structures or building heights which don't then suit a " square box" approach that is only efficient and cost effective when turned out on a production line. Simple changes to the standard product like lifting the windows to align with those of the surrounding existing buildings or introducing a feature that is non typical can make the product less cost effective than a traditional build.
Changes in planning in the coming years will ease this and we may then see a take off with modular builds. I for one welcome it as it might actually provide local authorities with a chance to supply actual cheap social rentable accomodation and remove people from the grip of unaffordable crippling landlord rentals.
Your roof is still going to need replacing.
Your boiler is still going to bed repairing or replacing.
You're still going to need a new kitchen/bathroom/boiler/roof at some point.
Those basic maintenance jobs are always going to be there.
And if there's no employment for you, how does one pay for these things? The market for everyone including those working with their hands is going to shrink and the trades will be scrabbling around for well enough paid, employed clients who can afford their prices.
Your roof is still going to need replacing.
Your boiler is still going to bed repairing or replacing.
You're still going to need a new kitchen/bathroom/boiler/roof at some point.
Those basic maintenance jobs are always going to be there.
And if there's no employment for you, how does one pay for these things? The market for everyone including those working with their hands is going to shrink and the trades will be scrabbling around for well enough paid, employed clients who can afford their prices.
Its even worse as the newly un-employed will be re-training as trades.
Governments not going to be happily paying out the new scroungers.
So you’ll end up with a saturated market of people hunting for work which will drop the rates for trades as well.
I'm feeling ok about it at the minute. I'm a videographer/photographer, all of my work involves real people talking on camera about real-world products used to create real music. If it's not that then I'm shooting live gigs. AI can't replicate that stuff, it can try and sometimes look convincing enough to my grandparents but that's about it, and why would it be used to try and recreate a live gig photo or video? The whole point of these images are about capturing a moment in time, not recreating it by what a computer thinks might have happened.
What is wrong with admitting you don't know?
AI simply amazes me. The certainty with which it will answer a question, and then do a 180.
"Does the DVD with the ean of 5036369800796 have a dts soundtrack?"
Surely the question has a definitive answer? And you would think a consistent one?
I have copied and pasted the question 5 times in succession, same AI, a full mix of "yes" and "no".
I presume the Internet contains varying reports about the disc, and therefore, the AI provides varying responses, and I get this, but it is the certainty of the response that gets me.
A response along the lines of...
There is not a consistent answer within my knowledge base.
Discogs say X
Redit says X and Y
Manufacturer says X
(I would trust the manufacturer [WINKYEYE])
Would be more helpful.
Grrr.
Spoiler, I'll give you my answer once it has dropped through my door!
What are you using. Claude Opus seems fairly comprehensive and reasoned.
Here’s the full answer with the reasoning laid out.
Identifying the disc. The EAN 5036369800796 resolves to Madness – At Madstock, a concert DVD of the band’s 1998 performance at Finsbury Park, London. The specific barcode corresponds to the 2006 Region 2 release, running roughly 75 minutes on a single disc. eBay The same barcode appears across multiple retailers (eBay, World of Books, Discogs, CeX), all pointing to this same title, so the identification is solid.
Finding the audio spec. Retail listings (eBay, PicClick, etc.) gave format, region, and tracklist but didn’t state the audio format, so those alone couldn’t answer the question. Discogs is the reliable source here because it catalogues technical specs per release. The catalogue entry documents the audio as Dolby Digital 2.0, Discogs alongside a 4:3 full screen picture and single-layer DVD5 format. Discogs
The conclusion. Dolby Digital 2.0 is a two-channel stereo track and an entirely different codec from DTS. So no, this DVD does not carry a DTS soundtrack.
A caveat worth stating. At Madstock exists in several editions (2003, 2004, 2006, 2010, plus an NTSC reissue), and I checked across them — none list a DTS option. They’re all consistently basic discs with stereo audio. This makes sense for the source material: a 1998 live concert recording released on a low-cost, no-frills disc, the kind of production that wouldn’t typically include a multichannel DTS mix. If a listing somewhere claimed DTS for this title, I’d treat it as a data-entry error rather than a genuine variant, given how uniform the specs are across every documented edition.
Is AI about to make you redundant?
No.
Not just a.i. The robots are now on their way.
What are you using. Claude Opus seems fairly comprehensive and reasoned.
Yeah, and that answer just cost you ten quid.😁
Meanwhile,
https://www.the-independent.com/tech/ford-ai-automation-human-workers-b3003787.html
Ford has had to rehire engineers when they found that AI automation wasn't up to snuff. It's cost them billions.
Thanks jam bo. Claud's answer seems pretty comprehensive.
May I ask, did it consistently answer in the same vein?
I'm using the AI built into opera's android browser. Maybe that is where I am going wrong? I have only assumed the A stood for 'artificial', if I were to read the small print, maybe I'd learn it actually stands for 'any-thing-but'!
Ford has had to rehire engineers when they found that AI automation wasn't up to snuff. It's cost them billions.
Although its unclear whether they did rehire the same people or just some competent leads who could manage a bunch of juniors whether AI or breathing.
I have been having "fun" recently with some specialist AI "prompt engineers". I would like to say they failed but that would imply they knew the target. I asked for a private meeting with them since I knew I was going to kick the shit out of them and so wanted to do it privately so before anyone else saw it they could correct it.
Instead because it was "AI" they doubled down on it and so I had to give a public kicking.
In many ways I see Ai as the same as outsourcing. Today it is your juniors but you need to plan for your next lead. Where are they coming from? Who owns your expertise? If you dont then you are ****ed.
I have only assumed the A stood for 'artificial'
There is a joke, although before the current llm approach to AI did well, that AI meant Always Indian.
Cory Doctorow says it'll be fine.
I think the tl;dr is that either society will find a way to reign in the excesses of AI, or we'll find ourselves in a revolution and burning it to the ground. At least I think that's what he said. So I'm not sure that what he means by "fine" is quite why I mean by it.
Still, he's always very entertaining.
The excellent Benn Jordan dives into a.i. in his latest expose of all things shitty, if you’re just after his a.i. rant it begins at 20mins in but worth watching all
What is wrong with admitting you don't know?
AI simply amazes me. The certainty with which it will answer a question, and then do a 180.
"Does the DVD with the ean of 5036369800796 have a dts soundtrack?"
Surely the question has a definitive answer? And you would think a consistent one?
The issue is it doesn't understand what its telling you - its offering you something that has the familiar patterns of an answer. Largely from ingesting conversations exactly like this one where all the contributors have varying viewpoints, experiences and knowledge.
Recently I was looking up intervals for replacing the timing belt on my van - its difficult to pin down definitive information as Stellantis don't put the information in the handbook or service book and don't publish it online (they also won't tell you if you ask them directly and tell you to ask a dealer instead). Its a van that has gone through a series of changes of engine over a short period so on forums where people have asked the same question lots of people answer with differing opinions because they are actually talking about completely different engines - and don't necessarily realise, the conflicting information isn't called out.
So when I search for the information Gemini's AI summary is that - There is no need to change timing belts on a Boxer van becuase the engine is chain driven and doesn't have a belt. Then in the second paragraph it details how often the belt (which it has definitively stated isn't there) needs to be changed. Even gives a part number for the belt.
It's not aware these are contradictory statements because the data it has harvested will pretty much always have those two contradictory statements being made authoritatively. So the answer it gives fits the pattern - but not summarised as 'some say' and 'others say' but as a statement of fact that both things are true.
You're holding it wrong.
A single LLM is not omnicient, its unfortunate that they attempt to be all the same.
The issue is it doesn't understand what its telling you
I have to constantly remind the people that make* decisions for their own business of this fact. One is certain that if he asks it to produce say - a sickness policy, that the output will comply with legal requirements.
*want to outsource
The issue is it doesn't understand what its telling you - its offering you something that has the familiar patterns of an answer
The bigger issue comes when you don’t understand what it’s telling you either.
It’s just a tool, and as such can be used well to craft amazing things, or badly to make a hot mess.
I use AI a lot in my job (education) and some of the things it can do are remarkable. It's allowed me to do lots of tasks in hours that would have taken me days or weeks in the past. So, the potential for improving productivity must be huge and that will presumably mean that fewer people are required to do the same amount of work. Whether that means fewer people employed, the same number employed but working fewer hours, or the same number working the same hours but producing more remains to be seen though. In the past it's always tended to be the latter I think.
The key point though is that it is only useful for tasks that you could do yourself. It can save a lot of time but I find that it rarely gives me the best result first time. If I understand the output, can spot the errors and weaknesses then I can prompt it to fix those. But once you start asking it to do things you don't understand you are in big trouble. Now you are no longer in control because you can't tell when it is lying to you. This is something I see all the time with my students. Although I'm tending to move away from assessing the output to assessing the process I still see lots of example of students trying to use it to produce output that they don't understand and then submitting stuff that is full of errors.
So, I'm excited by the potential but worried by the faith that some people seem to have in the results and where that might lead us. In terms of the original question; no idea but thankfully I'm close enough to retirement not to care 😀
The key point though is that it is only useful for tasks that you could do yourself. It can save a lot of time but I find that it rarely gives me the best result first time. If I understand the output, can spot the errors and weaknesses then I can prompt it to fix those. But once you start asking it to do things you don't understand you are in big trouble. Now you are no longer in control because you can't tell when it is lying to you. This is something I see all the time with my students. Although I'm tending to move away from assessing the output to assessing the process I still see lots of example of students trying to use it to produce output that they don't understand and then submitting stuff that is full of errors.
The trouble is now finding source material that isn't corrupted by AI. I was researching something fairly specific this morning (could flame detectors detect a flame that wasn't producing CO2, I know the answer is "no, but...." because they look for the specific infrared emissions spectrum of CO2 molecules, but there appear to be whole wikipedia style websites filled with chemistry and engineering junk that reads like it was almost certainly written by ChatGPT.
Today's AI 'fun'.
I uploaded a set of documents some of which are new, some old, and some taken from a regulators website. I asked AI to check these documents and compare with some due diligence expectations, some procedural cross-checks and repeated information.
Instead of doing that AI chose to tell me off for being a fraud as the boss was not me and the chair of my board was not who I had put down. It even gave the names of the people who it thought were the boss and chair - one is our founder (never been chair) and one is a name that I have never heard of...
All delivered with huge confidence and clear set of statements that I did not know my own job title or the person who chairs our board...
AI = increasingly shit?
AI was the final kick I needed to get off Facebook. I was mostly on there for interest groups and I mentally filtered out all the far right indoctrination that floated past.
One of the most useful groups was one for the camera I own. Lots of useful set up info and tips from users. But increasingly the group got flooded with questions like "I'm going on safari, what's the best lens to take". It was clearly AI bots scraping info.
So people started feeding rubbish back like "The best lens is the Sigma 25-1000 f2.8" a lens which doesn't, and almost certainly won't ever exist.
So now if you ask AI what the best lens is to take on safari, it confidently tells you to buy a lens which doesn't exist.
Pointless waste of resources.
I use AI a lot in my job (education) and some of the things it can do are remarkable. It's allowed me to do lots of tasks in hours that would have taken me days or weeks in the past. So, the potential for improving productivity must be huge…
Can you give a couple of examples of these labor saving tasks?
I’ve been repeatedly disappointed at the time I appear to have wasted prompting ‘pro’ versions of Gemini and CoPilot (researcher, analyst) to, for example, mine a document corpus and produce useful results. Things were better with I2E and TERMite
There are certainly things it does well and things it doesn't do as well and my uses are obviously very specific to my needs, but things I've found useful in teaching include:
Creating a graphical user interface to illustrate some concepts that I was teaching. Basically an app with buttons and sliders to change various inputs and show the effect on the output for the process I was teaching. Generated around a thousand lines of python code in a few hours. The first attempts were not perfect and it's an example of how it is only really useful for things that you can check. There were a few cases where the inputs didn't produce the correct outputs but it was quick and easy to just tell the AI that this bit was wrong and get it to correct it. I could have written the code myself and have written similar code myself in multiple languages over the years. But it would have taken me days or weeks to produce what even the free version of Claude managed in an afternoon.
Generating slides, notes, figures or electronic textbooks for some introductory courses. Areas of basic science where there isn't really any debate work quite well.
Generating multiple sets of data for assessments (to reduce copying) or materials for group exercises.
That's just a few examples, but hopefully you get what I mean.
Thanks. That sounds similar to my experiences. On the whole, not good enough for serious use but OK for quick and dirty or novelty value. Unless I’ve misinterpreted.
