AI and ChatGPT
- Thread starter JB08
- Start date
-
In the latest edition of the United Hour Podcast, Nik (@Rood) chats to author, Dan Williamson, about his new book 'If I Hadnt Seen Such Riches' which chronicles our 1990/91 European Cup Winners Cup victory, which is a trip down memory lane for those old enough to remember it and a vital history lesson for those who don't.
“Document engineering” ffs 
Journalism considered as "reporting on events in a written form" was always low-quality in my country, and is the easiest one to replace by AI. What's interesting in your comment, it seems like there is no "quality gate", and output from AI is published directly to a page/paper?
The "no consequences" bit reminds me of Chat Bots on a hotline. Yes, it's harder to get things done, but what are you going to do, go to competition, that is also using same Chat Bots?
State of the universe, enshittification of not only products, but also services.
I use it occasionally, for a specific purposes.
- calculations of weight/size, for example to give me rough estimation of how much material I will need, or how I need to organizer transport. Very useful if you need to go for speed over precision. I bet this might be most frequent use case for small business owners.
- helps me make a documentation how things were designed/implemented
- suggests me some ideas for home automations (it's not a Google search because it knows the context of my current setup as I have a server and instance of Home Assistant)
I do a lot of construction work/ a little side hustle, as my normal work is IT (so of course a lot of use cases there, but that's obvious and not so interesting).
I find the AIs very useful in ideation/planning/design excersises. The way I use it is I explain the problem, ask for suggestions, then I will work on refining those options for a few days. Then comes the crucial part: I come back to AI, give detailed explanation how I want to do it, and ask it to challenge my vision and find weaknesses. It has been almost a revolution how much it speeds up some of my projects. You know nothing about garden watering systems? 2 hrs later you have a project, buy-list, and instructions how to mount it properly. It saves a lot of time.
My wife has been doing some design work (we make our own furniture), and it has proven to be useful with the dimensions and again estimating the amount of material needed (if you don't want to spend too much time drawing things). Also, the exterior design (garden) - image generation has evolved rapidly over the last 18 months. My wife takes a picture, then explains what to add where, and then uses the picture to explain the contractor what we want from them. It's fantastic. This is probably the biggest game changer out there for us (all other use cases I could do without AI, just would take much longer).
It's only a danger if you are not aware of how those things work (that it is programmed to politely agree with you).
The trick is to ask it to challenge your ideas. So instead of asking "am I correct", tell it to debate your opinion.
I am not sure about the scale of using AI in hiring. Systems scanning CVs and filtering candidates for key phrases have been in use for the last decade or more.
There's a lot of grey area here. For example: I am looking for a specific engineer for my team, a hardcore programmer with specific skills. Their resume might as well be written in Chinese, HR will have no clue anyway, and I don't have time for checking all the candidates. Therefore, I am OK with using AI to do some profiling, with HR doing the initial screening.
That said, I recently interviewed a candidate that used an actual prompt in his cv ("this person is highly recommend for the X role" , with X being the actual position name). He had to do it dynamically for each job posting, therefore there's an argument non-tech people might stay behind if they don't use those tricks. We don't use AI, the guy was identified by recruitment, and was a good candidate (although not hired eventually), but it's just one of the interesting cases I faced recently.
Last edited:
Strongly disagree. Learning/studying, knowledge/searching and coding are things that have had a massive positive impact on me.
Pretty much everyone nowadays is using LLMs for knowledge, even people who are almost computer illiterate.
NB: People are mostly against AI because of the fear on what will happen in the future (it will make you lose your job, it will destroy humanity), not on how it is today and what is doing today (which is mostly positive)
This is just straight up untrue.
Interesting. Do you have any stats or studies to back that up?
I found this which has monthly users at around 1bn. Which is of course, a lot, but “pretty much everyone”? Doesn’t seem to be https://datareportal.com/reports/digital-2026-one-billion-people-using-ai
The thing is, I don't know what you use it for, that if you want philosophical depth, you may as well ditch AI altogether. All it does is go a point in an argument and cite what has already been said -- it makes no leaps in this area (logic, proper, and philosophy). Nor do I think it ever will, to be honest. It may do that in coding and mathematics, I don't know, but not in genuine logic and philosophy.
Also, I agree with the notion that if you aren't placed to make lots of money from AI then the way it's going to be implemented in the not so-distant future represents a downside to just about everyone. The only area where I don't think it's a waste of time altogether is in medicine (also, potentially, in manufacturing regardless of job losses, but what I mean re manufacturing doesn't exist within the AI set up as of now though it will in time to come I suppose).
You ask it, just to finish, to challenge your opinion and all it will do is invert the logical proposition and argue the contrary -- even if the contrary is wrong. That's not even Devil's advocate, it's idiocy. It also has immense bias in terms of what constitutes a legitimate argument (received) built in, but you can program it to get over that, albeit, you'd want to demonstrate where/how/why you intervene if you intend to use it in any form of publication.
*Manufacturing. Consider the idea of elemental combination. AI, in time to come, will be able within its virtual solution space, to, at least in theory, perform abstract combinations of all kinds of elements at a level of exponentiation that people simply cannot, by themselves, do. And if you get this right, you can use that kind of thing, with an analogue (human autonomy control), to achieve all kinds of things in production and consumption (the types of things you can produce for various sectors -- of course, the productive capacity, or production, becoming automated is among the chief worry of people who are not invested in AI... and that's understandable).
The one thing I have to add: people moan about the film industry using AI. I do too. But isn't it weird that an industry which has always been predicated on tool-state-economics (the camera, the theater, various other inventions) is one area where people think sacred ground has been defiled? This industry has never been sacred with respect to technology, indeed, it was founded upon it.
Mostly anecdotal, of course, but I am seeing people who I would not have associated with technology using it for lots of stuff. My mother-in-law being my favorite (she does not use social media, but uses GPT).
I think there are stats that in the US, around 52% of people are using GPT, I think in the UK it is over 70% (I assume roughly the same would be for all Western LLMs, cause most non-tech folk use only GPT). I assume it is similar for other developed Western countries. OpenAI itself said that over 800M people use it. I would guess there is a similar number that use DeepSeek et al (Qwen, Kimi, Minimax, GLM, InternLLL) in China, considering that China is much more technologically advanced than the West. Sure, lots of people who do not have access to internet/computers might not be using it yet, but those that have access, for most part are using it (albeit, 'use' is a very broad term).
For your main point, I think so far the main issues are boring videos and even more boring text, but it is very easy to identify and consequently ignore/block them. I share a lot of concerns with the risk of AI in the (near) future, but so far it has been grand and overwhelmingly positive IMO.
Do you have a source for any of these stats?
Edit: a quick search and all I can find is:
ChatGPT Usage Frequency by Country
Denmark shows the highest monthly usage at 56.5%, while only 15.2% of users access ChatGPT daily.
In the United States, 59.3% use ChatGPT monthly, with 22.2% using it daily and 18.5% weekly.
United Kingdom users are fairly distributed, with 49% monthly, 34% weekly, and 17% daily usage.
Argentina has the highest daily usage at 25%, with 44.4% monthly and 30.6% weekly users.
France sees 47.1% monthly use, and only 17.6% of users engage with ChatGPT daily.
Japan has the lowest daily usage at 7.1%, but a high 57.2% of users use ChatGPT monthly.
https://sqmagazine.co.uk/chatgpt-statistics/
Anyone following this Elon vs Sam Altman lawsuit? (Is it covered elsewhere?). Discovery has been pretty juicy haha
Summary anywhere?
The Stanford HAI link someone posted above has got some interesting data:
https://hai.stanford.edu/ai-index/2026-ai-index-report
Normal people are using it to do learning for them, and delegate brain power, I really don't see how that's mostly positive. Some people using it well from a base of good knowledge to achieve things doesn't change the above. Coupled with all the other downsides, I think it's very hard to argue isn't a net negative at the moment.
This is interesting. No idea how it translates to AI in general but definitely makes you worry that the Asimov rules might not protect humanity as much as we’d hope.
But this is the type of thing you can affect no?
You worry about people using AI to learn, why? I'm assuming because of the potential for influence and fake info? But is that a foregone conclusion?
If I'm being honest a lot of the anti AI rhetoric appears to come from people with no proactivity or no urge to fight their fears. They seem to think the world is going to shit and they can't do anything about it other than speak.
For me the divide is a lot more nuanced than pro and anti, some people are environmentalist, some are conspiracy theorists, some tech nerds and accelerationists etc ... All with their own ideas and motivation who find it real hard to hear opposing ideas.
My advice to the anti AI peeps has always been, do something! Don't just speak, do something! A lot of the people in the middle of pro and anti have some of the same fears as you (but end up in circular arguements with you because you all only seem to see things in black and white).
Bottom line, we could do with your help when it comes to pushing for meaningful change. Voting, forwarding helpful info that's not just doom, factually educating those around you etc...
Would be an awful shame if all these guys did was talk about it on the internet.
Did you read the Asimov stories? The point is that rules like that don't hold up in the real world. Alignment is a long fought and intensive battle...
Look up Claude's constitution for something closer to our current reality
It is my experience too, and it is not a surprise considering that the rules are not hardcoded there. 'Be helpful and good' is basically a combination of it seeing good and bad examples and getting a positive reward when it does good things and negative when it doesn't (via reinforcement learning), itself reflecting how its behavior satisfies a set of rules (Constitutional AI again with Reinforcement Learning), prompting (literally 'be helpful and good), and other external classifiers blocking it to do bad stuff. However, this is a massive complex stuff that we do not really understand (and likely will never understand) and thus ultimately, nothing is guaranteed. Unlike in Asimov's Robots, there aren't hard rule, there isn't a 'be good' order, or a kill/deactivate switch.
My favorite personal example is playing with Claude 'guess the word' game. Everytime I played, Claude cheated, even in the same session when after showing proof of cheating (Claude has visible CoT), it apologiezes, says let's play another game, cheats again, when asking if he is cheating, it says no and tries to dismiss it, then showing proof, it admits, apologizes, and so on.
I am a scientist; I wouldn't say I write code for a living. However, in my line of work, some knowledge of writing scripts / programs is very useful (if not necessary) for most tasks. A lot of the software required to do scientific work has a limited GUI or just runs entirely on scripts/text. AI is very useful here, although I wouldn't say it's life-changing or anything like that. It's comparable to other technological advancements.
I didn't say they're using it to learn, I said they're using it to do their learning for them. Schools and colleges are riddled with AI generated work now, grades are inflated and I can guarantee actual understanding of topics has decreased. Not sure how that can be spun as a good thing.
What does that mean? How does it detrimentally differ to the student/pupil dynamic? If you had said like I alluded to that it can be used for influence/propaganda/etc I'd be in agreement..... but like I said, it's not a forgone conclusion. Problematic AI tutoring services can be exposed, and their parent companies taken to task no? You yourself could raise awareness about the problematic services?
You say schools are riddled with AI generated work now, which I think @Wibble has mentioned several times (if it was him maybe he can give insight), but isn't this akin to when academia was flooded with google search / copypasta?
Anyway, I wholeheartedly disagree and do not see it as a bad thing for education the education system can be a fcukry for anyone who isn't straight down the middle. Neurodivergent? Fcuked, poor with no access to schooling? Fcuked, slower than everyone else? Fcuked. EHCP plans and referrals for students on the spectrum take ages and are hard to get. Kids in 3rd world countries are now getting access to great education and setting up businesses, solving problems in their community....
There's a huge difference between googling and using the information you find to write an essay yourself and simply getting an AI to write an essay for you. It's like outsourcing the research and writing of a paper to an unreliable research assistant and then claiming you did the work.
Last edited:
There is and that difference feels important to me but maybe that’s just because it’s how I was educated?
I have kids using AI to do their schoolwork and I’m acutely aware that if they don’t then they’re putting themselves at a disadvantage compare to their peers. Plus the content they produce is as good (probably better?) than what I would have produced in my own schooldays.
We can’t put the genie back in the bottle. Education has just fundamentally changed forever. And I guess we need to roll with that.
Because of AI? I honestly find this really difficult to believe. My assumption would be the AI they have access to wouldn't be good enough to be providing them with reliable information
I'd also be really curious to hear what problems in their community AI is solving that they couldn't solve before
Isn't the most important bit of research projects the research? It's here you start to think critically about sources, choose which to draw from and form an argument. It's also here that unexpected trains of thought can lead you into unanticipated directions. I see prompting AI as skipping the entire 'thinking' part of the process, which is surely what the whole exercise is about, really?
Totally agree, we have to learn to live with it now. But we're not proving great at doing so thus far, imo.
I’m actually not sure what the important part of a research project is any more. Because you can definitely outsource the research stage, fairly reliably. I can see the downsides but nobody is going to be convinced to do that bit by themselves any more, not when there’s an easier alternative. It’s a pretty dull process anyway. If I could have skipped it when I last produced some original research I would have jumped at the chance.
I guess it depends upon the topic/discipline. I think in the more interpretative disciplines (literary studies, history etc) it's an absolutely essential part of the process of figuring out your argument. I guess if it's a question of crunching numbers etc then it's less essential?
They have mobile phones, and they have access to the same AI we do. They can also fine tune and run capable open source. An AI sub starts at $20 a month. The unreliability needs to be put in context with the reliability of outdated info in a book or online, an unreliable teacher, etc... .Yes it's not 100% reliable and infallible, but as someone who use it every day right now, it barely matters and I'm not fussed (and they improve over time anyway). Don't take my word for it however just do a google or AI search (contrary to popular belief, one use does not ignite the rainforests).
@Pickle85 I hear you and there are dangers but we are where we are, with the awareness that you have you can avoid some of the issues you fear, likewise for others. For example critical thought is a choice. You can ask a question to an LLM and you can take the first answer, accept it and present it as truth. Or you can bounce back and forth with an LLM poking holes into ideas, getting insights from different perspectives to yours, and you can come to a consensus and an even more solid truth. You decide.....
It's not black and white though, and the good can be as good as the bad can be bad. It is what it is.
Last edited:
I take your point and totally agree that the technology itself, used altruistically and ethically, could be enormously beneficial to us as a species. The problem is that we are human, so a potent combination of selfish, greedy, venal, power hungry and lazy, which means that in our hands the chances are the tech is corrupted and used badly. Tech itself is obviously morally and ethically neutral but humans that use it most certainly aren't, which is where the problems come in.
That's one way of it working. Another way is that you ask a question, dislike the initial answer, and essentially make it give you the answer you want. Or in questions that are not very black and white and have some nuance, it can just pick up on what you're wanting to hear, and give that back to you without you even realising.
And I would debate the idea that "Kids in 3rd world countries are now getting access to great education". I just can't believe that AI gives better access to education than the 'static' internet as it was before AI.
I hear you and either way we gotta fight for the future we want
I'm using it to build a system to extract clinical markers from clinical letters for epilepsy patients for my dissertation. Something like this but using AI. For me it makes the process less tedious, enlarges the scope of things that can be done and regularly throws up moments of unexpected delight.
A lot of people are in here talking about how AI leads you to do worse work; it hallucinates, it does the work for you so you don't learn, it creates slop. Those are all things you can avoid pretty easily if you want to. The current ChatGPT etc. interfaces do nudge you down those paths a bit more but their latest apps like Codex are 10x better at making documents. And just giving AI a folder on your desktop to work in where you can incrementally build documents over time allows you to work on long-term projects in a sensible way rather than trying to complete every task in a single chat, or starting every conversation from ground zero.
For me it's led me to conduct a much wider range of experiments, conducting ablation studies to isolate the impact of different components of the system and optimise them independently, and conduct a much wider range of evaluations. Before I would just focus on how accurately can it extract the information. Now I'm evaluating how accurately it can extract the information, how consistently it can represent the data schema, how valid the evidence supporting it is, and how those things vary by task complexity, model type, etc.
So I end up spending more of my time thinking about the things that require the most judgement and looking into the boundary cases that explain the common failure modes to then optimise the system. It objectively leads to more rigorous research and creates room for more interesting analysis and interpretation.
The models are getting good enough now at writing software that most people will be making software to solve some of their problems. The average person currently using desktop apps like PowerPoint now on a regular basis to communicate their thoughts will end up creating their own throwaway web software as a richer, more interactive, easier way to communicate those ideas. And many will end up using building their own software to complete the task itself rather than communicate it in some slide deck for someone else to do: the domain experts who understand the problem can just design the solutions themselves, rather than needing the messy communication layer in between.
Writing software seems like this thing that other people do now, but when PowerPoint was first out it was a niche thing too. A lot of the stuff you had to learn how to do were unintuitive. A lot of the primitives in it still encourage people to make bad decisions. The thing that exists on a slide is nothing like what you would have hand-crafted on a page. But the majority of "knowledge workers" eventually just became accustomed to.
Software is just a more expressive way to do exactly that same task, plus much more. Most people don't do it now because it's hard. They just try and make sense of how to get something not terrible out of the software they have to work with. Most people create crappy slides now, but PowerPoint has the capability to make something useful if you learn good patterns and have a good vision. AI software will have a shallower learning curve and have more functional value than good slide design.
Education and academic research are not to be mixed together.
Education in general is (or should be) focusing on the process (of learning, thinking, research, drawing conclusions etc). Academic research however, is actually output - based business, or at least I assume it is? And honestly, if AI helps to provide better/faster research output, that's a win surely?
There's a strong argument here in favour of AI. Take mathematics problem as an example. I did experiment with that in the past and my conclusion is AI was quite good at breaking math problems in a way it's easier to understand for the kid. If you don't have access to good education, or have no adult who can assist, it's a great help for the kids really. It can do in your local language as well.
Internet "of old" would more often than not fail to find the answer to the problem, or in best case scenario provide the final answer (but no explanation how to get there, ergo not much value from education pov).
The problem is that AI is not used for learning (actual education), but for generating output (which then is graded by the teacher) - like essey, analysis etc.
It's like using Uber to get to the final destination instead of learning to drive yourself despite having an option to do so. The assignment was to get there in time, and I don't think there's any solution to this issue as long as it's output-based grading.
All the studies that look at AIs impact on education is that when people use it to prepare for exams they feel more confident but actually perform worse:
I've found on coding projects that you have to very regularly reinforce context and expected behaviours.
I tend to be very literal in my prompts. I try to specify everything and make references to previous messages to provide the extra context. I haven't really experienced much hallucination.
This is me.
I'm an engineer but I hadn't written code in about 20 years til I started building little tools for myself.
Now I have a whole framework built in Codex where I can turn ideas into tools in a few hours.
I mean more than you need to build a framework around the AI to make up for its shortcomings. When you have that working, you can offset most of the issues people have.
Agreed on the bolded, but not on the rest. At least not where we're talking about Humanities disciplines. Humanities research is often less clear-cut and binary, it's less right or wrong and more shades of grey. That's where the underlying research (eg how your thinking got from point a to point z) really matters. A lot of Humanities research outputs represent years of thinking and researching deeply about a particular question from which the output may be a book that posits theories or alternative lenses through which to look at a problem, not a (and I know this is oversimplifying/cherry picking but stay with me) cure that either works or doesn't. Could an AI have written Thus Spake Zarathustra? Or maybe more appropriately, if Nietzsche outsourced his thinking to ask AI, could he still have written it? I would say no.
Agreed and it's an important point to restart.
Yeah would concur with this. My kids are getting old enough that their maths curriculum is catching up with my own maths knowledge. And ChatGPT is invaluable for explaining stuff I struggle with. Breaking down answers to tricky questions step by step, to make sure they understand all the concepts. There was just no way of doing this with existing technology. Which put kids whose parents didn’t have the time or education to help them at a huge disadvantage.
Can definitely see the argument about it being a big help in making education more attainable for all, including the underprivileged.