AI.... awesome. Really need it. It'll raise productivity and enable a new middle class of fulfilled, happy people.
Over to JohnV, the other one...
AI will eventually find its way into virtually everything. That it's only in a couple of applications now and we can't say with certainty what's next doesn't mean it won't find its way. Compare to the ubiquitous chip / integrated circuit, the inventors and early investors had no way to predict the massive uptake and the huge variety of devices that have only come about because of the ability to package circuits onto tiny silicone wafer.
The real issue is that in today's connected society 'everyone's' aware of the rise of AI and the markets are so connected they can overheat as we see based purely on the potential. When the IC was invented (1959, had to google it) I'm pretty sure no-one was clamouring to buy shares and create start ups to make them. I don't honestly see AI any different really to those big innovations, and its not whether it'll become the size expected, it's whether it'll do it fast enough to please the markets.
🙄
.
.
.
.
.
.
.
. 🤣
Laugh if you want, when ai and qc start working in tandem - and some put timescales of 3-5 years on big steps in that - it might be niche again to start with but the impact will be huge.
https://cetas.turing.ac.uk/publications/ai-quantum-computing-and-high-performance-computing
https://www.nature.com/articles/s41467-025-65836-3
It's not a matter of if but when, and the AI market will make money, it's just whether it's fast enough for the investors liking.
I keep reading it as Al as in AL not AI as in ai.
Good bloke that Al works bloody hard.
I keep reading it as Al as in AL not AI as in ai.
Good bloke that Al works bloody hard.
As they are my initials, I can confirm this to be true /cough
Ai, from an investor point of view, is a bubble.
It will change the world, but not to the scale that's being pushed, if it were, it would usher in the end of capitalism and even capitalists must have a limit to their greed when it means their own demise.
Like dot.com the bubble will burst, there will be a nasty crash and the money men will lose interest and go looking for the nest get rich quick thing.
I keep reading it as Al as in AL not AI as in ai.
My name is Alan. Whilst I don't go by "AL" some people use it anyway. Reading AI as a capital i tool some brain retraining. I started writing it as A.I. in the hope it'd catch on, but to no avail.
I know it's hot and I slow down too in the heat, but this was 7am before it got hot - this morning's AI classic when looking for a small error in a result:
Me:
You're completely right — I made an error. 7.3874° rounded to three decimal places is7.387°, and to two decimal places is7.39°. There's no valid rounding that gives 7.393°. I simply made a mistake in my earlier working. Apologies for the confusion.
And there was the error, a rounding error of the daftest sort.
To be fair it has also generated code to resolve a question that was based on a transcendental equation. I didn't know what that even was until I realised why layers of Excel calcs could never do what I was trying to do. And AI got it all right in the end. Yet it can't round reliably .. Fascinating.
‘AI + quantum computing in 3 to 5 years’. Sounds like the kind of claims made 3 to 5 years ago, again. Maybe I’ll be laughing on the other side of my face when our quantum robot overlords take power in 2029-2031 I doubt it.
I remember claims being made for productivity increases from the introduction of computers, office ‘productivity’ software, ML, ‘big data’, etc. lots of changes, for sure, but more productivity? Just different productivity ime. I suppose it makes a lot of sense for an exec team to cut PAs, admin, IT/IS as it cuts overheads. But IDK if having to juggle my own meeting scheduling has done anything but remove my PA’s jobs and make less time for doing the work I’m meant to.
What is definition of productivity? Take new drug discovery, for example banks of chemists designing and synthesising new molecules they think will have therapeutic value. Most fail; I've a mate that has spent 35 years in discovery led teams, and so far has one success that was his (OK, now management and team has more but of his own work). AI and QC is going to revolutionise.
The 3-5 may be optimistic, but it will happen in time. Like I said, whether investors remain patient or not is a different question.
I'm sure it'll transform certain areas, but mainly it'll just be this...
I'm at stage 6 already
AI and QC is going to revolutionise.
yeah, right. AI has promised to revolutionise drug discovery and development several times in my decades in the industry. Some improvements to some activities is the result. The decision-making process for what to hold in the portfolio remains human for good or ill.
With all respects, your decades in the industry haven't coincided with the availability of scaleable fault tolerant quantum computers either. That plus AI is, I believe, what will turn this from missed promises to reality.
Do you think it'll never happen? Or are we just arguing over how long? If the former - then I'll say that you're wrong. If the latter, then instead of saying what hasn't happened yet - how long do you think it'll take?
Like everything in this messed up world - it will be at the service of money-men (and to that end it will implode) rather than good public outcomes.
This current iteration (entertaining stupid people on the net) will probably die and come back - like much technology - in a different form.
The sooner that happens the better.
Things are aligning for a pretty rough economic ride.
This current iteration (entertaining stupid people on the net)
I'll use AI daily. But never for entertaining stupid people on the internet.
It's actually been incredibly useful at work and has saved an awful lot of time.
I write things and edit other people's writing for a living. Sometimes that involves more thought, sometimes less. I mostly write (and edit) technology-related stuff. I was an IT journalist for a long old time, and worked (and was laid off- yay!) through the dot com boom and bust. A lot of the bullshit I saw then from tech executives is being echoed now by AI execs. You may have noticed that the crypto bros pivoted to being AI bros almost overnight, and there's a reason for that: it's a more profitable grift. When you see them pivoting back to crypto then you know the gig is up.
ChatGPT was supposed to replace writers first. I can confirm it has created lots more work for us to correct, as the writing is universally as dry as dust and oddly repetitive. It's also replaced a whole bunch of quite basic and repetitive copywriting (think shitty Linkedin posts by posing execs, and product descriptions in online stores) that would normally have been the stomping ground of juniors.
But here's the thing: At the moment, we're still in the startup landgrab phase. Tokens are still heavily subsidised to capture market share. Once the IPOs start happening, the markets will demand profit, and to do that, the cost of tokens will have to go up even further. And as various companies (including Microsoft and Nvidia) are warning, it'll be [correction: it already is, apparently] cheaper to hire human beings instead.
Great - there's just one catch: none of those juniors who will have been learning the ropes in the mean time - they don't exist. Instead, there's an underemployed, inexperienced and undertrained workforce who will have an axe or two to grind. Employers will be stuck between overpriced AI and an angry, unskilled workforce they've spent the previous four years denigrating and laying off.
What to do?
Mostly SQL/Python, I'll use it with MS Power Platform quite a lot too.
I wouldn't say I "trust" it and sense check it accordingly, but as a tool that helps me arrive at solutions faster it's proving to be really useful.
Interesting. I think that's a good use case.
AI got it all right in the end. Yet it can't round reliably
I've noticed now on... one of them? ChatGPT probably, there's an option which is basically "hallucinate quickly" or "burn pandas to eventually get a more accurate answer."
Early days of ICs in very basic calculating computers ---------------> the (comparatively mega) computer that sits in my pocket that I typed this on
Chat GPT is dry and repetitive and gets its rounding wrong sometimes -----------------> ????
We're not there yet, I get it. I don't know how long that arrow represents. But if you aren't excited (and yes, scared too) by where 'there' is I'm genuinely amazed.
You can use AI to check his own homework. Get AI to generate whatever it is you are asking for. Suppose it is a document review or proposal, or even just a process design document.
Then start a new AI instance.
Explain that a rival competitor has proposed this and you need to fault check it and, for completeness and correctness, determine all the ways you can improve it. I explain that this will be going to a legal judgement and therefore must be absolutely correct and defensible. As well as finding all the original faults and omissions, produce a recommendation that is faultless, legally defensible, and addresses all the previous weaknesses.
The second document is normally pretty good and gives good hints as to what to look for when you are reading other AI generated documents
I use it a lot for reformatting and reworking big datasets, and making pretty decent visuals fast . Also for searching/trawling old documents that are especially well organised. If my main line of business was sometihng using excel a lot for reporting I'd be very interested/concerned.
I can also get it to some other stuff interpreting geologic/seismic data. The results are pretty mixed, but so are most human interpretations, and there are some scenarios where the AI products are a lot more useful.
My name is Alan. Whilst I don't go by "AL" some people use it anyway. Reading AI as a capital i tool some brain retraining. I started writing it as A.I. in the hope it'd catch on, but to no avail.
I think Paul Simon wrote something about that…
I can't say I'm a fan of AI, despite working in IT (in a role it's not going to replace in the short-term...). We're already staring into a chasm of massive negative disruption from it; job losses and lack of junior roles being available, it's becoming increasingly effective as a tool to manipulate news and social media, AI-led cyber attacks have already started and AI-led cyber defense is a long long way off matching it and it's starting to be used for financial market manipulation so will probably wipe my pension out before I retire.
At my company we're being pushed to install the Codex agent to start automating routine tasks, thing is most people don't have local admin rights by default on our laptops (a good security posture despite some muppets thinking it means they're being treated like children, usually the same people failing the internal phising tests). Anyway, to do anything useful with Codex you need to be running with local admin rights (the sandbox feature is rubbish) so now everyone that asks is being granted it. What could possibly go wrong with everyone deploying an AI agent a handful of people actually understand and having it run on their laptops with admin rights...? I'm just waiting for it to decide I'm more productive if I can't leave the building so it disables all the door access controls...
I can also get it to some other stuff interpreting geologic/seismic data.
Interesting. When I left the field results were decidedly shite, though it has some obvious attractions - handling more data than a human can possibly scroll through. Upgrading existing fault mapping tools (“ant tracking”) looks like an obvious place to start.
On a less niche topic - Tony Bliar seems to have drunk gallons of AI KoolAid, and believes it’s the answer to every problem.
On a less niche topic - Tony Bliar seems to have drunk gallons of AI KoolAid, and believes it’s the answer to every problem.
That’s because he’s being paid handsomely to do so. Blair’s unwelcome contribution yesterday should have been prefixed with “and now a word from my sponsors…”
My line manager loves it, as he has not worked out that it is not an actual person, and thinks that its complementary language is just specific to him and that he is now a genius. I can no longer talk to him, without first filtering his stuff through AI and returning my answers back in the same format. It's getting proper odd. He also uses it for his city breaks, to find places "off the beaten path" - he's no longer human but an interface. He's lost the ability to think or write in sentences, or remember what was just discussed. Sadly he is probably the future, where people no longer make any decisions for themselves.
Won’t AI plus QC make IT security non-existent and therefore IT systems unusable for anything critical - from traffic lights and energy / power systems to defence assets - because someone will be able to hack in at will unless your AIQC is bigger than theirs, which in turn means the power drain will make it impractical?
Beware intelligent hardworking engineers with no concept of wider consequences.
Just a thought.
We’re not really talking about “A. I.” Here are we? We’re talking about LLMs aren’t we.
which are fine as far as they go, but it’s just another way of gathering and processing data, to achieve a user defined outcome. The more complex the requirements the bigger the potential errors (IME). It’s inherently a technology you can’t trust in critical (safety related) applications.
The amount of resources being poured into”AI” is kind of staggering, the money (and circular financing), the data centres, the hoovering up of hardware manufacturing capacity… It’s never going to deliver on the claims, that’s how the bubble bursts ultimately.
just like with the .com bubble a handful of useful technologies and companies will survive and reap the benefits over the following 20 years.
And just like the personal computer and internet the promises of improved work/life balance and productivity will not come to pass, you’ll still work the same hours with no reduction in stress, and become a slave playing keep-up with a box full of GPUs in a data centre somewhere…
We’re not really talking about “A. I.” Here are we? We’re talking about LLMs aren’t we.
I always thought that LLMs were a form of AI, but I'm not sure I've got my definitions right either.
I tend to think of AI as being some sort of algorithm (machine learning, deep learning etc) that takes inputs and tries to predict an output. There will be some underlying structure (e.g. a convolution neural network) and the parameters for that structure (e.g. weights) are calculated by training the model with lots of data (supervised or unsupervised). So, it could be looking at a medical image and trying to predict whether it is normal or abnormal or it could be taking in a number of tokens (words or parts of words) and trying to predict the most likely next token. But it's all just different types of AI.
Also, I don't think the frontier models most people refer to are strictly just LLMs these days are they? I think those models have all sorts of other algorithms wrapped around the LLM (including, hopefully, safety guardrails).
Take the example earlier of the rounding error. That makes perfect sense for a pure LLM. It's not doing any maths, just trying to predict the most likely next token. So, if it's seen the string "7.393" near to the string "7.3874" and the string "rounded" in its training data then it may say that's the most likely result. But I think you can now tell most frontier models to look for maths and when it finds an equation to generate a bit of code to solve it rather than just using the LLM.
Whatever the terminology though, these models are having a profound effect in my area (university education). There is a lot of focus on how students can use them to render lots of existing assessments meaningless but there is also huge scope to use them to speed up all sorts of tasks, from curriculum design to lesson planning to generating those assessments. That's before we even get into the thorny issue of whether they could or should be used for marking and feedback.
Is it just a bubble that will burst and we'll all go back to the old methods? That seems unlikely now. Will it reach a plateau? Maybe. Will it continue to grow and improve, so this is just the tip of the iceberg? Who knows? If I did I wouldn't be wasting my time typing this 😀
Some days I think it's an amazing time to be alive and working. On other days I've glad that I'm getting close to retirement.
Won’t AI plus QC make IT security non-existent and therefore IT systems unusable for anything critical
Most current cryptography is based on very difficult sums and QC is very good at deciphering difficult sums. But quantum cryptography relies on physics, and is thought to be unbreakable. Of course, it's an escalating conflict and staying ahead is hard, but significant effort is going into it.
https://www.nist.gov/cybersecurity-and-privacy/what-quantum-cryptography
On the power - QC is way more energy efficient than classical. A superconducting QC, generally the most expensive type in terms of power usage uses around 25kW for a 1000 qubit processor. And a fraction of that is in the QC itself, most is in running the fridge that the QC is housed in. And while 1000 qubits is impressive, they can fit way more into the fridge so larger QCs will broadly use the same power - it's not physical size that's the limit, it's other technical issues.
Other types like trapped ion or atoms, or photons are less advanced but likely to be substantially less, as they can be operated at room temperature as only eg: the atoms are being cooled and not by thermodynamics per se, but by a process known as laser cooling. (in a circular process, some of my colleagues are using quantum techniques to understand new materials that will help reduce this power burden further)
Interesting. When I left the field results were decidedly shite
How long ago? There's plenty cases where results aren't shite, and they're getting better all the time. The example I often use because colleagues were part of it was in breast cancer studies - training an AI system to scan mammograms for early signs of breast cancer, and then set it to work on a load of old images. Example trial in the link, the system achieved in a couple of days what it would take a radiologist a working lifetime to do in volume, and spotted the suspect scans for further review just as well as the radiographer. However, it also flagged some false positives - not good, leads to worry and potentially further invasive testing, etc.
EXCEPT - turns out they weren't false. Mia was spotting early stage cancers that radiographers couldn't find, leading to earlier diagnoses = better outcomes. If you think that's a bit meh based on your own experience of LLMs, I think you're missing the point.
https://www.bbc.co.uk/news/technology-68607059
AI's not perfect, but part of that is 'deliberate'. Like teaching schoolkids maths, you don't only set them easy sums as tests. You set things they get wrong so you can find out WHY they are getting them wrong, and then teach them on those things. Another example (and yes, I know civil liberties and how the authorities use the data, etc.) is facial recognition. Early trials got things wrong and particularly had issues with dark skin colours. So the systems have had extra learning and recent trials in Croydon had 1 false positive in 470,000 scans. 0.02%
"Compare to the ubiquitous chip / integrated circuit"
Why? Seems like an arbitrary comparison. You could equally compare it to the myriad technologies that failed to deliver on the hype. I think this is an example of survivorship bias or something.
The only impact on my industry so far has been a proliferation of bots trying to harvest information so they can vibe code software solutions to problems that either don't exist, or where another piece of software is a completely illogical solution.
That and people using it to draft emails without realising how obvious it is that they've used it, consequently making themselves sound absolutely stupid and losing all credibility. Oh, and making industry LinkedIn even less readable than before.
We’re not really talking about “A. I.” Here are we? We’re talking about LLMs aren’t we.
which are fine as far as they go, but it’s just another way of gathering and processing data, to achieve a user defined outcome. The more complex the requirements the bigger the potential errors (IME). It’s inherently a technology you can’t trust in critical (safety related) applications.
Yes, LLM not thinking machines.
"Compare to the ubiquitous chip / integrated circuit"
Why? Seems like an arbitrary comparison.
Fair criticism, I'd say not totally arbitrary as it's all about physics and microelectronics really. I guess it does have the potential to fail to deliver, personally I don't think it will, as per previous post it already is. I was more making the point that looking at the earliest incarnations and dismissing it as a bit shit before it's really developed - what would those correspondants have made of the Wright Brothers, who achieved about 50 yards 10 feet off the ground? "Well that was underwhelming, I'm putting all my money into hydrogen balloons"
AI-led cyber attacks have already started and AI-led cyber defense is a long long way off matching it
It hasn't merely "already started," it's here in spades. The number 1 point of initial breech has been variations on credential compromise - mostly phishing - for years. It's recently been supplanted by vulnerability exploits and this is certainly a direct result of AI tooling.
Early trials got things wrong... So the systems have had extra learning
I think this is the thing a lot of people miss. As someone once said, "AI today is the worst it will ever be."
I ran a query through ChatGPT yesterday, then out of curiosity threw the same thing at Copilot. ChatGPT's output was good - unnervingly good in fact - whereas Copilot's was the sort of terrible result where it didn't understand that words were more than random lines and shapes. It struck me that this 'slop' is what we were all mocking as the best AI could do, really not all that long ago.
AI is getting better. And it is getting better fast.
ChatGPT's output was good - unnervingly good in fact - whereas Copilot's was the sort of terrible result where it didn't understand that words were more than random lines and shapes.
Which is weird really as CoPilot supposedly uses the same OpenAI models as ChatGPT. Or at least it can if you've selected them.
Interesting. When I left the field results were decidedly shite
How long ago? There's plenty cases where results aren't shite, and they're getting better all the time.
About 3 years ago. If you have more recent examples from seismic, I'd be curious to see. DM me if you don't want to fill the thread with something that may be OT for most people.
If you think that's a bit meh based on your own experience of LLMs, I think you're missing the point.
My comment was about seismic interpretation, not general.
It's not possible to be across everything that 1000+ scientists are doing (especially the stuff they don't talk about) so I only tend to hear about stuff that gets reported through more generic internal and external channels, so I can't claim to know anything specifically and would be googling. I am aware of some very interesting work on seismic detection but not specifically AI for that purpose (although it is quantum phenomena that is used) and I'm pretty sure that AI will be involved in the analysis of the masses of data that is generated.
[as we're getting involved time for my disclaimer...views on the future potential of AI and QC are my own, not any official position of the company. They are informed by what I do in Science and Engineering, and the discussions I'm involved in as part of my day job, but all that info I'm citing or sharing is chosen because it's in the public domain. And although I'm a scientist by trade, I'm a/ not directly doing any of this research and b/ not a patch on the intellect and capability of my colleagues that are. If anyone is also in the trade and wants to be put in touch with the true experts, happy to arrange that]
To steal and mangle someone else's line: AI success is unevenly distributed. It works really well for some things, not so well for others. There's no doubt it's a transformative technology, but it's been massively overhyped by people trying to grab as much resource and real estate as possible before the inevitable drop. Just this morning, the 'AI will take your job' cobblers got rowed back by OpenAI.
Over a decade ago I was working through what my employer was doing with machine learning, data science (remember that?) predictive and prescriptive analytics, and figuring our what benefits it brought to users. I have to remind my current clients that AI in cyber isn't anything new - they and everyone else have been leaning on ML for years.
I get that it's getting better fast - just like every other early tech does and has done for a while. But it's benefits and impact are currently massively overblown and overhyped to attract investment. For it to be sustainable, it has to be profitable, and for it to be profitable, it has to be commercially viable. This is the crunch point: IPO. Have a rummage through OpenAI's S-1 filing for a few insights.
For balance, I asked Chatgpt what it thought and if AI will 'raise productivity and enable a new middle class of fulfilled, happy people.' Here's what it said:
This was the most interesting bit of the ramble above
- Human value shifts toward judgment, relationships, taste, leadership, creativity, coordination, and domain expertise.
Fortunately, I'm no longer employed in the corporate world so much of the above goes straight over my head. It's quite likely going to be like Dotcom where 99% of AI businesses are going to fail, whilst a few will reap the rewards. Phenomenal amounts of cash and resources will be consumed in the process.
I run a small design and print business and we sell into our local tourist market - making stuff to sell in shops plus a little bit online. What I have noticed is the sheer volume of AI generated rubbish that now dominates online marketplaces like Etsy - if I search for the name of our town/village I get pages of AI generate pictures/posters. The problem is they bear no relation to the actual place - buildings and whole streets where none exists. We had a situation in a shop I work where our supplier had created a new wine list, with an 'image' of the town - only it was AI generated slop. We sent it back and asked for a 'real' image to be used instead.
It's quite likely going to be like Dotcom where 99% of AI businesses are going to fail, whilst a few will reap the rewards. Phenomenal amounts of cash and resources will be consumed in the process.
All the players know this. They all want to be Amazon - ie the last one standing after the crash. All throughout the .com era people were saying Amazon was the biggest basket case due to its incredible levels of spending / investment.
