|
Welcome to a brand of
Mathematical
Services. 2026-08-30 13:00:09 - Paul D. Foy - ArtificialIntelligence Okay looking through my ramblings, it doesn't appear that I've made a post on AI and the consequences of all these AI developments, about which we hear interminably something on the news here and there. What's my take on AI? This explosion of digital phenomena of computing things: firstly it is clear that there has been a large explosion and that there has been a step change in developments. I'm thinking particularly about the large language models, which have obviously been developed with data over a period of time but they are making a revolutionary change in that they're granting access and knowledge effortlessly to almost anybody. You can direct your researchers on the particular subject as you like and the content is fairly good and you regard it as authoritative. On my specific discipline, mathematics and software engineering, it's a somewhat nuanced picture, I think, and it's not quite as I thought it might be initially. I think actually there is even more of a need for software engineers and I don't think software engineers will be made redundant very soon at all. There's more need for knowledge of software things and software techniques and digital techniques and how to navigate this landscape, even more than there used to be, to tell the truth. I must admit it doesn't make it very easy for old people, I don't think, here, including my mother. I've experience of that. On the ability of AI to generate code, it's true it can very well generate code and it can provide routines for you which you can slot in but they're somewhat beguiling and somewhat misleading. For example you can't actually use these routines often as libraries or as things you fully understand. They might be doing things in unexpected ways. On the whole it's very difficult building from them unless you understand them very well and you understand exactly what they're doing and that takes a bit of effort. It takes the same amount of effort as it traditionally did to code, inspect things, and to understand somebody else's code so there's no time saving there. Generally it is my experience. This is my experience. I'm speaking from experience. I'm speaking from experience of trying to use these AI-generated routines, these ChatGPT-generated routines, into my own code and to use them without too much amendment. It's not possible because they don't necessarily do things in the way you expect or the way you want. The black boxes of which you don't know the innards often and you can't rely on them to do exactly what you want to do. There's no time saving actually. As a time-saving technique, I find. It hasn't proved all that useful and I'm surprised about this. I thought they'd be more useful but there is a certain amount of use in the coding things of Claude and ChatGPT but it's not quite as I expected. They're not quite as easy to use as I expected. At the end of the day you've got to fall back on the human pace of doing things often, I find. So this AI way of generating code is actually, I'm thinking, very much like Matlab was regarded when I was working as a software engineer before. There you could knock up something, you could get something visually displayed, and you could see it all but you had no idea how it was doing it or how you'd have to do it yourself to do the same thing. I think it's strange. It's put the prototyping back into the software engineer rather than anybody else to tell the truth. It's making the software engineer even more important than anybody else. That's quite a quirky development I think. All these things have got to be evaluated and used and appreciated on their merits, I find. Perhaps it's what the developers of these tools wanted. They wanted more power for the software engineer, the poor hapless software engineer. Well I think it's wored from that point of view. 2026-10-17 09:58:48 - Paul D. Foy - I'm having a lot of thoughts about AI (artificial intelligence) these days. It's quite fashionable to think about AI and I'm having surprising revelations and understandings of things because of AI. For example, image enhancement: you often find people touting that they're using AI to do image enhancement, to do this and that with their photographs. How is this actually working? How is AI doing it? How does an individual verify that the photograph is actually sharper and is a better photograph? How is it done? It's got to be a human that judges that, hasn't it? He's got to verify that it's sharper. I don't think, as I understand it, AI can come along, create some kind of algorithm, and make it sharper, and then the AI says to it, "Yes it's now sharper. " I think AI is working by comparing and contrasting lots of images of similar scenes, a massive database of images of similar scenes, and sharpening it like that. I don't think it's creating an algorithm. It's trying an algorithm to make it sharper. It's doing it from the past experience of similar scenes. This is very much how AI works. It works on the past experience of similar things or by dredging through past utterances, the past similar things, the past scenes, this kind of thing. This is how AI works. Now I'm coming across this because I've been developing software to make impressions of photographs and things like that. I've found that when I've been doing this software, I've been doing things with the image and I've been making it worse. It makes it, to me, look worse than the original photograph. There might be a branch across a roof and it creates a notch in the roof line. I don't like that, that kind of thing. That's it. That's how my computer algorithm is working and I sometimes don't like it. It doesn't look good. It got me thinking: what's AI? What's AI? How does that do it? It's just some kind of algorithm, is it? No it's not. It's comparing lots of different scenes. That's how it's doing it. What I think is very interesting is to get into what the algorithm is that makes the scene pleasurable to the human eye. What is it about it? What is the nature of the image? What is the nature of the bits comprising the image, the pixels comprising the image, that make it pleasurable to the eye? That's not a question AI is answering. AI is answering the question: "Well I've done this before. I've seen this before so it must be like this. We'll have to do this. " That's how I think AI is doing it. I'd like to put developments into seeing what the core ingredients of seeing it are. To tell the truth I'm not sure that's traditional AI. I have to say here that every single word up to this point was written using Wispr Flow, developed by their own AI. I must admit it's incredibly good because I was umming and aching, stopping and stuttering, and all this kind of thing. It sorts everything out to make a nice flowing structure. 2026-10-13 02:09:20 - Paul D. Foy - I think these AI tools, ChatGPT, Claude, these kinds of things, they're actually putting power back into the individual and to what it means to make a personal contribution, your own contribution. For example I'll say what I mean. Many times in my career I've been told, "Well you can do this, you can do this quickly. Just use this and that to do it. You should be working a bit quicker than that. He, Joe Blogs over there, he can do it quicker. He's already done it before," to which I replied, "Oh well, that may be very well but you've asked me to do this. You want my comment on it. You want me to verify that it's correct, to understand what I'm doing it for myself. You want my contribution here. " I think AI is just enforcing that in a way. It may well be that AI can blurt out code and solve things and structure everything and do this and that but if there's been no human verification of it, if it's not passed the human test, if a human hasn't verified that it is true, if a human hasn't had some input into it, what worth is it? Who's it being done for? How do you know it's correct? There's got to be some personal involvement in verifying that things are correct. There's got to be the human saying, "Yes this is good stuff, this is good work, this will work well. I can understand this. " There's got to be some human involvement in here and I think AI is forcing this viewpoint and forcing this to occur now: for the code review to be done, for the approach to be structured, and for it to be shown that it is a good approach, that things are working well, and that there's been adequate contemplation of things. I think AI is forcing this more than ever so I still view it as a good thing in a way. 2026-09-23 19:04:40 - Paul D. Foy - I think you can view Claude and ChatGPT as other ways of replacing another part of the software development cycle, that is, as a code reviewer. I must admit it: Claude, for example, comments on and reviews my code and suggests improvements and design flaws and design weaknesses quite well. It's very good at that. It can't just write reasonable code, which you've got to spend some time to get to grips with. To tell the truth it's not all that easy getting to grips with what it writes because it spouts it out quite effusively. As a code reviewer it's quite helpful. I'd say the software engineer now is doing all jobs: - the project manager - the code reviewer - the designer - the mathematician He's got everything under his bonnet. The poor software engineer. Everybody's becoming a software engineer!. 2026-08-30 19:39:46 - Paul D. Foy - Yes I'm seeing these tools, Claude and ChatGPT, as prototyping tools predominantly. They're pretty good at that. They're prototyping for the software engineer. Perhaps Matlab is prototyping for the mathematician, even for the software engineer, but the software engineer can now do more with these tools so that's quite good. I quite like them for that thing. I can't really rely on them for doing serious things, to tell the truth. Doing something that has to be correct at all costs and can't have any bugs in it, they don't operate like that at the moment. Perhaps they may do, perhaps things will improve. It'd be nice to speak to the people behind these algorithms, these LLM models and things like that, what they're actually doing. I must admit I don't fully understand the algorithms of the core of it. They do seem to be using someone else's work that has been done before. They're scooping up from somewhere a routine that has been written down, done before. In some sense I don't think they're originally thinking and creating the routine there. They're patching things together from what has been done before. I can tell this because there are often mistakes in them, mistakes as a developer might make anyway, and they've found those mistakes in somebody's work somewhere. I see them as prototyping tools at the moment, to tell the truth. They're good ones, they're well worth having. I can quite see that Claude and ChatGPT, though, might be making some kind of offering for the student market, for the academic community, much like Matlab was successfully doing. I can see that coming along. Not sure academics need to be too worried, certainly not in the software engineering and mathematics field, more so the software engineering field because it is just an explosion of things to do and an explosion of software engineers needed. Software engineers are needed more than ever it seems to me. Other disciplines, well, it's replacing all knowledge. All knowledge is inaccessible at the tip of a fingertip, at the tip of a well-structured question, so that's difficult for other subjects. I think if you're a historian or a medic you can just ask the question. It'll give you all the answers with sources often. We're all becoming computer scientists!. Post a comment: |