Chat GPT

[Nano]

Sure, it may not be as efficient or as clean code, as a guy on 50k a year, spending two months on it, but it can get 90% there with an hour of back and forth, which is basically how my coder mates use it, and they use it for apps, websites, databases etc, they all love it. It's getting better all the time too of course. Clean code or code which is the most efficient as it can be is largely overrated too mind, for the vast majority of uses anyway. No point trying to write the most efficient thing ever and taking 6 months on it, when you could have something functional in a week, and then spend the next 5 months making it better. It's just another very good string to the bow, which when combined with other resources just makes everything better/ more efficient.

I suppose it's best used with a minder, but also opens the door to amateurs to crate basic functional programs with very little experience or knowledge.

Whether it is true AI or not doesn't really matter to most, if it acts like it does etc.

99% of people don't understand how the world works, and they never will, maybe nobody ever has.

I am a competent self-taught SQL coder, even did a role as a sysadmin temporarily, and I always shared your opinion that if it works it works and that'll do. After working alongside people that had studied and understood the reasons that efficiency is important I realised I was just ignorant because I was seeing things from an end-user point where everything is working fine until it isn't. In a one-person to one-server environment you push your buttons and if it works then great but in a virtual environment with 100s or 1000s of concurrent users then efficiency makes a huge difference.

A good analogy is simple traffic flow. If you have a 4 lane motorway with a speed limit of 70 and 1 user going 50 then it doesn't matter. The user will travel from a to b. If you have 4 people using it. it still doesn't matter. There are 4 lanes. Put 100 on and those people driving slowly really get in the way.

If you push something out that is good enough to get going and then start fixing problems then you are doing roadworks and we all know what sort of traffic problems they cause. People get much angrier about something not working the way it should than it not being ready yet so it's worth getting it right first time.

 

[Glover_elbow]

It will get to the point (it's probably already there now) where you can learn 10 subjects enough with AI help in the time it takes you to learn 1. You don't even need to remember everything, just scratch the basics, learn and dump and move on, all the simple things just get filled in over time and you can re-learn them back at any time.

There's probably not much point in learning every nook and cranny in a topic if you have the best reference tool by your side to explain something, and it gets better at explaining it every year. You can basically spend more time on the key harder bits that you will need to know, it will basically just save masses of time.

It will probably end up being allowed to do kids homework (or be the end of homework), that should be the aim of it, and those who get the most out of it should get the best grades. It's a waste of time doing homework on basic tasks which anyone could do with AI help etc, and it should allow more weighting to be put on actual in-person tests, without the use of AI.

I think it will transform education for the better in time. Rather than thinking about it giving answers, it will be like each kid in the class having the best teacher and best library beside them. It wont be long before a child home schooled with ai is outperforming traditional schooling. Not that i am advocating missing the social interaction of school.

 

[Laughing]

Sure, it may not be as efficient or as clean code, as a guy on 50k a year, spending two months on it, but it can get 90% there with an hour of back and forth, which is basically how my coder mates use it, and they use it for apps, websites, databases etc, they all love it. It's getting better all the time too of course. Clean code or code which is the most efficient as it can be is largely overrated too mind, for the vast majority of uses anyway. No point trying to write the most efficient thing ever and taking 6 months on it, when you could have something functional in a week, and then spend the next 5 months making it better. It's just another very good string to the bow, which when combined with other resources just makes everything better/ more efficient.

I suppose it's best used with a minder, but also opens the door to amateurs to crate basic functional programs with very little experience or knowledge.

Whether it is true AI or not doesn't really matter to most, if it acts like it does etc.

99% of people don't understand how the world works, and they never will, maybe nobody ever has.

For very good reason, we are not allowed to use it for coding at work. It really isn't very good. As a coder if you can't do the basics like connect to and render data from a REST API, you shouldn't be anywhere near a proffessional coding environment. 35 years in the industry makes it fairly clear that it writes poor code, outside of boiler plate suggestions.

We are a country mile away from automated coding. Rational Rose tried something similar 22 years ago and it suffered from the same issues. Yup, it could write code, just not very good code and that was based on very complex UML diagrams so has a very clear idea of what the coder was trying to achieve.

Your comment about most people not understanding how the world works is flippant and not very accurate.

I assume everyone on this board, for example, understands how mfc relates to the town. ChatGPT doesn't. It doesn't even understand what football is, nor what the town of middlesbrough is.

I do this for a living Andy, with a lot of experience. It's fun to play with and might manage to do a year 9's homework. It's not very good beyond that.

Some real world applications of ChatGPT are usually around search, so you can get contextual results using natural language and for completing metadata on things like written articles for indexation on a search platform like elastic search or coveo.

It is currrently, vey limited in the industry, despite, was it Altman, spouting in the press last week?

We are a million miles away from general AI simply because we don't know how it works in nature. AI can't even generalise on basic concepts.

I'll say it again, ChatGPT looks impressive because it is a natural language platform and as humans we athropomorphize speach, which we really shouldn't.

Interestingly enough, the really clever stuff doesn't get the publicity it should because it's a bit boring. Expert systems still make up the vast majority of AI systems. They are fantastic at recognizing cancer markers, for example. It doesn't have the pazzazz though of a chat bot.

 

[Andy_W]

For very good reason, we are not allowed to use it for coding at work. It really isn't very good. As a coder if you can't do the basics like connect to and render data from a REST API, you shouldn't be anywhere near a proffessional coding environment. 35 years in the industry makes it fairly clear that it writes poor code, outside of boiler plate suggestions.

We are a country mile away from automated coding. Rational Rose tried something similar 22 years ago and it suffered from the same issues. Yup, it could write code, just not very good code and that was based on very complex UML diagrams so has a very clear idea of what the coder was trying to achieve.

Your comment about most people not understanding how the world works is flippant and not very accurate.

I assume everyone on this board, for example, understands how mfc relates to the town. ChatGPT doesn't. It doesn't even understand what football is, nor what the town of middlesbrough is.

I do this for a living Andy, with a lot of experience. It's fun to play with and might manage to do a year 9's homework. It's not very good beyond that.

Some real world applications of ChatGPT are usually around search, so you can get contextual results using natural language and for completing metadata on things like written articles for indexation on a search platform like elastic search or coveo.

It is currrently, vey limited in the industry, despite, was it Altman, spouting in the press last week?

We are a million miles away from general AI simply because we don't know how it works in nature. AI can't even generalise on basic concepts.

I'll say it again, ChatGPT looks impressive because it is a natural language platform and as humans we athropomorphize speach, which we really shouldn't.

Interestingly enough, the really clever stuff doesn't get the publicity it should because it's a bit boring. Expert systems still make up the vast majority of AI systems. They are fantastic at recognizing cancer markers, for example. It doesn't have the pazzazz though of a chat bot.

Obviously, you're not going to use it off the bat, especially not for anything complex, and nobody who knows what they're on about is saying you would, but it can do some fairly complicated things extremely quickly, and then you or it can refine as needed. It needs someone with the ability and experience to oversee this though, or spot what needs refining, not something which a novice could do very well etc. I must know about 30 people who basically code all day, every day, and have done for the last 20 years and they all say it's excellent, with the right controls/ supervision. The one who I've spoken to the most about it says it's like having an assistant who he can get to do the donkey work, and it's easier and quicker to get it to do what he wants rather than asking a trainee with A levels and a degree.

Just becuase something was tried 22 years ago and didn't work well does not mean it can't work well now, or won't work soon, especially in the tech industry. Every year AI is put to the test against coders and ranks better and better, and even by 2022 it was scoring in the top 50% I think, I bet 20 years ago it was bottom 5%.

For coding, it will only get better now that the world has been opened up to it even more, and it's being tested more and more and they're ironing out problems more and more.

Most people don't understand how the world works, most don't even understand how their own country works or why it works. 52% of the UK just voted for brexit and their eductation is probably in the top 10% of the world. AI doesn't need to understand what football is, but it can tell you what it is. It can already watch the game and build a stat database of that, and analyse it better (and much faster) than any human could. It could combine/ comb the knowlege of more than any human ever could ever learn.

You're right about the clever things, analysing, modelling, crunching masses of data and varying scenario's etc. ChatGPT is just like a shop window which kind of shows it to the world, in whatever they're interested in, but will end up good enough to basically be almost anyones asssistant. Still relatively early days and even Chat GPT has moved on a lot over the last 7 months.

It is just going to get better at everything, and very quickly, like it has done over the last 10 years, just look at the examples below, and apply that to almost any field and it will be or can be similar.

 

[Andy_W]

I think it will transform education for the better in time. Rather than thinking about it giving answers, it will be like each kid in the class having the best teacher and best library beside them. It wont be long before a child home schooled with ai is outperforming traditional schooling. Not that i am advocating missing the social interaction of school.

Totally agree, you could learn most academic things pretty well with youtube, academic resources and chatGPT already, enough to get a solid footing for a job which could earn 100k a year, or use it to re-train in something else.

You could also have 100 kids learning at their own pace, rather than all going along at the same lesson pace, and just have a teacher to oversee it, but I bet it can already beat most teachers with assistance/ help, it's already beat doctors for compassion and quality of an answer for 80% of the time.

 

[Laughing]

Obviously, you're not going to use it off the bat, especially not for anything complex, and nobody who knows what they're on about is saying you would, but it can do some fairly complicated things extremely quickly, and then you or it can refine as needed. It needs someone with the ability and experience to oversee this though, or spot what needs refining, not something which a novice could do very well etc. I must know about 30 people who basically code all day, every day, and have done for the last 20 years and they all say it's excellent, with the right controls/ supervision. The one who I've spoken to the most about it says it's like having an assistant who he can get to do the donkey work, and it's easier and quicker to get it to do what he wants rather than asking a trainee with A levels and a degree.

Just becuase something was tried 22 years ago and didn't work well does not mean it can't work well now, or won't work soon, especially in the tech industry. Every year AI is put to the test against coders and ranks better and better, and even by 2022 it was scoring in the top 50% I think, I bet 20 years ago it was bottom 5%.

For coding, it will only get better now that the world has been opened up to it even more, and it's being tested more and more and they're ironing out problems more and more.

Most people don't understand how the world works, most don't even understand how their own country works or why it works. 52% of the UK just voted for brexit and their eductation is probably in the top 10% of the world. AI doesn't need to understand what football is, but it can tell you what it is. It can already watch the game and build a stat database of that, and analyse it better (and much faster) than any human could. It could combine/ comb the knowlege of more than any human ever could ever learn.

You're right about the clever things, analysing, modelling, crunching masses of data and varying scenario's etc. ChatGPT is just like a shop window which kind of shows it to the world, in whatever they're interested in, but will end up good enough to basically be almost anyones asssistant. Still relatively early days and even Chat GPT has moved on a lot over the last 7 months.

It is just going to get better at everything, and very quickly, like it has done over the last 10 years, just look at the examples below, and apply that to almost any field and it will be or can be similar.

View attachment 58149

View attachment 58148

I tried. I can't be bothered explaining what is wrong with your post Andy. Have a good one

 

[Cuthbert]

ChatGPT writes decent bioler plate code, no more than that. It is very poor at writing efficient code, real-time code and anything that requires specialist knowledge. There are better resources at the minute for learning or indeed stealing code. It isn't a replacement for github.

Bing is slightly better and has access to the internet.

I am not even sure ChatGPT can be called an AI. It understands natural language and does a good job at fooling you into thinking you are taking to something sentient.

To be fair, it did pass the turing test, but most computer scientists don't think this is a good test.
As I said above, it doesn't understand the world ot how it works, despite having a huge index, nor will it, in my lifetime.

Could it write some code like yours for football predictions?

 

[Andy_W]

I tried. I can't be bothered explaining what is wrong with your post Andy. Have a good one

You explained some parts as your opinion to counter something I never even said, and some of the rest was pretty much the exact opposite of 30 other professionals who I actually know (and have known for 20-30 years).

 

[Laughing]

Could it write some code like yours for football predictions?

Yes, it could, that is pretty mundane stuff. It couldn't do much in the way of advising what data points to use though.

A basic deepnet is fairly easy to write as a programming exerrcise.

 

[Laughing]

You explained some parts as your opinion to counter something I never even said, and some of the rest was pretty much the exact opposite of 30 other professionals who I actually know (and have known for 20-30 years).

But openAI agree with me, that it can produce boiler plate code and not complex code. It sounds like your coding mates arent very good coders Andy. I think even chatgtp would agree that it can't write anythong other than simple, boiler plate code. Ask it to write some threaded code for a 4 core cpu and see how efficiently it discards those threads?

As a single example, it will wait until all 4 threads have completed execution before discarding any of them, pretty much the same as an inexperienced coder would do, if you ,multiply that by thousands of threads, say for example reading a switching panel 24/7 where recycling threads is necessary, in real time, it leaks memory and crashes.

It writes boiler plate code and that is pretty much it. It is worthless in real world problem solving.

Have a good night, I have better things to do than argue with someone who clearly knows very little about writing code.

 

[Andy_W]

I am a competent self-taught SQL coder, even did a role as a sysadmin temporarily, and I always shared your opinion that if it works it works and that'll do. After working alongside people that had studied and understood the reasons that efficiency is important I realised I was just ignorant because I was seeing things from an end-user point where everything is working fine until it isn't. In a one-person to one-server environment you push your buttons and if it works then great but in a virtual environment with 100s or 1000s of concurrent users then efficiency makes a huge difference.

A good analogy is simple traffic flow. If you have a 4 lane motorway with a speed limit of 70 and 1 user going 50 then it doesn't matter. The user will travel from a to b. If you have 4 people using it. it still doesn't matter. There are 4 lanes. Put 100 on and those people driving slowly really get in the way.

If you push something out that is good enough to get going and then start fixing problems then you are doing roadworks and we all know what sort of traffic problems they cause. People get much angrier about something not working the way it should than it not being ready yet so it's worth getting it right first time.

Funny you should say that, as one of the lads I've chatted to about this is a sysadmin I think, or high up in his company and he's basically the database guru. I spoke to him specifically as I needed to use MySQL in a limited way and remembered he had been using SQL a long time, and had been to college and uni doing that sort of thing etc. When I first started asking him questions (obviously very low for his level) the first thing he said was why get ChatGPT to write what I needed and then go from there. It did what I intended (better than I originally could), and I got it to explain what it was doing and then just refined this to suit. Did the job well enough for what I needed, then I moved onto to something else I needed, and to be honest I've already forgotten most of was needed at the start.

Sure, for a complex system, which has people's lives on the line, you're going to want to make things as efficient and bug-free as possible, but let's not pretend that all programs written by coders are efficient or bug-free, as they're not.

It's like with gaming etc, in the old days games had to have as little number of bugs as possible on release, as there was no way to patch them or make them better after they got shipped out. Now if they waited for the time taken to stress test everything nothing would ever get released on a yearly basis, so games just get pushed out basically as beta's and then patched as bugs get reported.

It's horses for courses though, some things need to be 100% off the bat, things like military code etc, but even then they never are 100%. I know this from when I worked on Eurofighter, as we were constantly doing updates and fixes which were not meant to be required, for faults which were apparently impossible, and we were doing them every week.

For some other things, getting a product at 80% in use immediately is better than waiting a year for something 90% of 95% etc, all depends on who the client is and what they want. Having 0% for a year is undesirable to anyone, but equally using a system which is bugged to hell and unreliable isn't much help to some either.

 

[locomansimon]

It's a terrifying prospect !

 

[Andy_W]

But openAI agree with me, that it can produce boiler plate code and not complex code. It sounds like your coding mates arent very good coders Andy. I think even chatgtp would agree that it can't write anythong other than simple, boiler plate code. Ask it to write some threaded code for a 4 core cpu and see how efficiently it discards those threads?

As a single example, it will wait until all 4 threads have completed execution before discarding any of them, pretty much the same as an inexperienced coder would do, if you ,multiply that by thousands of threads, say for example reading a switching panel 24/7 where recycling threads is necessary, in real time, it leaks memory and crashes.

It writes boiler plate code and that is pretty much it. It is worthless in real world problem solving.

Have a good night, I have better things to do than argue with someone who clearly knows very little about writing code.

Nobody is saying it can write complex code, with supreme efficiency or accuracy, or change the world off the bat? My mates bosses seem to think they're good coders and they're all on good money and seem to get raises and promotions often, so I reckon they're ok.

Nobody is saying it currently has the ability to replace an experienced coder, but you've practically proved my point that it can do the job of an inexperienced coder or someone with little to no knowledge altogether, and it does this already.

It may be "worthless" in your eyes, or for your use, but it's not to millions, which is why millions are using it, and why Codex and Copilot etc are being developed.

You're arguing for your use, and saying that only your use exists or has any use, which is clearly not the case. It's kind of like saying that only professional websites exist, when it's not the case and anyone could knock up something which serves their purpose with little expense or knowledge.

 

[Laughing]

Nobody is saying it can write complex code, with supreme efficiency or accuracy, or change the world off the bat? My mates bosses seem to think they're good coders and they're all on good money and seem to get raises and promotions often, so I reckon they're ok.

Nobody is saying it currently has the ability to replace an experienced coder, but you've practically proved my point that it can do the job of an inexperienced coder or someone with little to no knowledge altogether, and it does this already.

It may be "worthless" in your eyes, or for your use, but it's not to millions, which is why millions are using it, and why Codex and Copilot etc are being developed.

You're arguing for your use, and saying that only your use exists or has any use, which is clearly not the case. It's kind of like saying that only professional websites exist, when it's not the case and anyone could knock up something which serves their purpose with little expense or knowledge.

It is really like talking to a psuedo expert on everything Andy, are you sue you are not a chatbot?

I am not arguing against using it for my use. I am arguing against using it for any coding that has any impact. It is often hopeless and can produce code that won't run, or worse still run, but not do what you want it to. It is hopeless. It can't even organize a project for VS creating stub classes. It's hopeless mate.

As for beginer coders, not sure about where your mates work, but we don't let a graduate anywhere near live code for about a year after graduating. Thats precisely because we write complex code. So no, I am not proving your point.

 

[Mickey_2003]

This thread sounds like Alexa and Cortana having a to and fro. How many times was ChatGP used to answer a question in here

 

[Emmersons_BrazillianDong]

I used it a few times a few months back, when it was using pre-2021 data, and it was ludicrous then, probably one of the most impressive things I've ever seen, or impressive in how things actually changed so quickly. Since then it's had an update and is now using current data I think. I've not even used Bing or Bard and they're meant to be even better in some ways.

I started re-learning programming before using chat GPT, and it seems beyond excellent at creating code (I think it uses codex which has been about for a while). It was far better than I ever was when I could actually code, back in the day, and a thousand times faster (and can do what I want in 100 different languages). With a bit of hand-holding, you can really do amazing things, but the people who will get the most out of it are the trained people and experts who can keep it on track, as it just saves a ludicrous amount of time. It would be hard for a novice to use it with code (instantly), as they wouldn't know what to ask it, and how to keep it on track. But the crazy thing is a novice can actually use it to learn, and for this, it is incredible when paired with other learning tools.

Once I get through my current to-do list, which is very long, I'm basically going to create a few programs to effectively replace what I do at work, or 75% of the time-intensive but relatively simple tasks, which will allow me more time to focus on the 25% where the money is actually made.

To me, for running a business, using AI will help me get ahead of the competition, in the same way that building a website and using Google AdWords both did back in the day.

Took me about 10 minutes of using it before I realised this is really going to change the world, and extremely quickly, it was like first using a search engine, but with 100x the power, and then knowing next year it's going to be 1000x, and so on. It's just going to improve the efficiency of anything which has any involvement in tech, which is just about anything.

What automations are you trying to do that covers most of your work bud?

 

[JackG]

Here's a useful directory of AI tools

 

[Andy_W]

Here's a useful directory of AI tools

There's some quite good posters on Twitter sharing new stuff every day, looks cool.

 

[Andy_W]

What automations are you trying to do that covers most of your work bud?

test - it stragely wouldn't let me reply in full with what I had wrote so split it down below.

 

[Andy_W]

What automations are you trying to do that covers most of your work bud?

Basically, just dealing with tender enquiries and getting a quick but comprehensive response out. A job enquiry might contain 20 fields for all options (maybe up to 3 at the start, but revisions up to 10 or so), and then each option might have say up to 20 installations and each of those has say 20-50 fields.

From that input I want to form various quotes (plucking info from separate sources), materials order email, a boat load of fairly complex calculations and then also a basic AutoCAD script to draw a basic version of it (later refine this when we win the work), and crate staged invoices etc. Maybe also fill in fields for a RAMS template too.

There's software to do bits of this, but the lack of control does my head in, and nothing can do all of what I want.