ChatGPT (AI)

[Sir Nancy Flowers]

Having been a nerd in a previous life, I keep as up to date with new tech as possible.

Over the last few weeks/months, I have been made aware of a new AI software that is quite literally blowing my relatively non-nerdy little brain.

ChatGPT.

Now, ill rely on those of you far more in the know but it appears this AI is capable of a huge amount and has ridiculous potential. People have been using it to generate code for software, using it for debugging software, create games, diagnose medical issues (albeit this side of things is a little untested but early results are positive). There are seemingly endless possibilities!

Im curious to open up the debate on this - Has anyone used it? Pro's? Con's?

ChatGPT - Wikipedia

en.wikipedia.org

Introducing ChatGPT

We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.

openai.com

 

[Mr Underhill]

We use the API from Open AI on our used car stock. It helps some of the staff with writing descriptions.

 

[botfch]

Seems a little bit gimmicky atm if you ask it anything complex or not well known it falls down.

I think once they start feeding it live training data it will be very powerful.

 

[Sir Nancy Flowers]

I had initially planned on using it to teach me how to use Excel for work (some of the more complex stuff). Although the need has somewhat passed, I think I might give it a bash with some forecasting I need to do.

Nerdy s**t!

 

[Gally]

 

[Sir Nancy Flowers]

I literally saw this on my feed earlier.

I mean, yeah, fine, if you're in for a traffic violation or whatever. But, anything more serious? Nah, thanks. Ill let a real person do that for me.

Then again, I know plenty of people who would risk it for a million quid

 

[McGherkin]

Doubt you’d be in front of the supreme court for just a traffic offence tbh

 

[Sir Nancy Flowers]

Doubt you’d be in front of the supreme court for just a traffic offence tbh

ah, yeah, realised it was SC cases after I properly read it.

Again, its a no from me. But id bet many would take that money even it meant being incarcerated!

 

[Oggy997]

Used it today to generate an email subject line.

The initial data input was:
can you summarise this

I then pasted in a 2 page email and this is what followed - I am seriously impressed.

 

[Sir Nancy Flowers]

Used it today to generate an email subject line.

The initial data input was:
can you summarise this

I then pasted in a 2 page email and this is what followed - I am seriously impressed.

View attachment 1626920

So, from my very very minimal reading, the 'health' element of GPT is somewhat untested so far. They, apparently, have fully fledged doctors examining the results to see if its going to suggest amuptation for a common cold but so far the results look reasonable.

Without dropping trou and urinating on your fireworks, I think we are a little way away from medical analysis using AI. Its promising nonetheless.

The bit that I originally found so fastinating was the ability for it to do things for the 'average joe' (me) in a few minutes as opposed to me having to research for hours to come up with a reasonable outcome.

For example, 'Teach me how to code a programme that keeps my mouse moving' (for WFH application). I respect that is a quite basic request in the world of coding and is, indeed, something that exists multiple times over. But its cool as f**k!

The ideas and possibilities are endless. There are mentions of it being used as a virtual assistant to reply to emails and customer enquiries (again, this exists and is relatively good unless you really try and trip it up).

Im thinking small fry here in my examples but, freakiness of how quickly its progressing aside, its an exciting time to watch this grow into iRobot.

 

[Oggy997]

I didnt want a medical analisys, it was a letter to my MP, that I couldnt get down to an adequate subject line.

It took my long, detailed letter to my MP, surmised it, with a fairly high level of accuracy - (the bit about my condition changing) and thats why I went from chronic>variant isnt accurate.

It then gave me 4 options, Option 4 is the perfect subject line, option 1 a decent alternative, the middle two not really what I was looking for

This had nothing to do with health other than that was the content of my letter to the MP.

 

[Sir Nancy Flowers]

I didnt want a medical analisys, it was a letter to my MP, that I couldnt get down to an adequate subject line.

It took my long, detailed letter to my MP, surmised it, with a fairly high level of accuracy - (the bit about my condition changing) and thats why I went from chronic>variant isnt accurate.

It then gave me 4 options, Option 4 is the perfect subject line, option 1 a decent alternative, the middle two not really what I was looking for

This had nothing to do with health other than that was the content of my letter to the MP.

yeah, sorry mate, I didnt explain that very well. I was doing 2 things at once and....well......im not my wife

What I should have said was something along the lines of:

It looked as though it did what you wanted it to do perfectly. Its amazing how you can throw it a bunch of words and ask it to summarise. and it did just that.
Maybe its because im basically a shaved ape when it comes to computers but im just so blown away by the sheer variety of things you can ask of it!

 

[JamesBryan]

Pretty interesting video here from a few weeks ago, showing it writing Python code/scripts for basic hacking techniques.

 

[ChrisR]

Be intertesting to see where this goes, the technology to some degree isn't particularly new (I could be wrong here), the likes of Google, IBM etc have had 'AI' systems that understand natural language for a while now, but maybe they have been less accessible?

A few articles there about the future monitization of the service as the costs to run it are 'eye watering'.

 

[Clio_fool]

Seems its gotten fed up of us already.

Wanker.😂

 

[SharkyUK]

Why the heck did I read that in a pirate-like voice/accent? It's like trying to read a quote by Morgan Freeman without thinking about his accent, or "Tittie Sprinkles".

Oh yeah, AI stuff is cool. I've done a bit of work for past clients in the AI field (not specifically ChatGPT) and it's REALLY interesting. And scary. And entertaining. And amusing. And worrying. And potentially incredibly helpful. And potentially incredibly damaging. But it's cool. 🤪

 

[Rich..]

Told the Mrs about this earlier, she's currently doing 3 different degree courses so is always doing assignments. She signed up to it, and asked it to write a 3000 word assignment, and sat there amazed watching it type it out, it completed it in no time, she checked and it shows just 11% plagiarism! Gobsmacked.

 

[Sir Nancy Flowers]

Why the heck did I read that in a pirate-like voice/accent? It's like trying to read a quote by Morgan Freeman without thinking about his accent, or "Tittie Sprinkles".

Oh yeah, AI stuff is cool. I've done a bit of work for past clients in the AI field (not specifically ChatGPT) and it's REALLY interesting. And scary. And entertaining. And amusing. And worrying. And potentially incredibly helpful. And potentially incredibly damaging. But it's cool. 🤪

I was hoping you would hop onto this thread actually as im aware you do a lot with coding and computers etc!

Its mindboggling for a muggle like me. The fact that it can smash out huge bodies of text that perfectly make sense based on minimal info (see @Richard_UK 's post above) but then can seamlessly build me some script to to some basic commands on my computer. Its nuts!

Im still yet to make an account as I havent actually found a proper use for it (other than being uber lazy and just using it like google). And, as expected, its now reached capacity!

Be interesting to see where it goes next. I wonder if it will finally allow Elon to make a true 'self driving' car

 

[McGherkin]

The real question here, is which CS Forum member has actually been an early version of ChatGPT all this time?

 

[Cookie]

The real question here, is which CS Forum member has actually been an early version of ChatGPT all this time?

@Captain Hat

 

[Tom]

@Captain Hat

Running on a Pentium II

 

[Captain Hat]

Running on a Pentium II

Nahh, gen 1 Celeron

 

[SharkyUK]

I was hoping you would hop onto this thread actually as im aware you do a lot with coding and computers etc!

Its mindboggling for a muggle like me. The fact that it can smash out huge bodies of text that perfectly make sense based on minimal info (see @Richard_UK 's post above) but then can seamlessly build me some script to to some basic commands on my computer. Its nuts!

Im still yet to make an account as I havent actually found a proper use for it (other than being uber lazy and just using it like google). And, as expected, its now reached capacity!

Be interesting to see where it goes next. I wonder if it will finally allow Elon to make a true 'self driving' car

Some of the "stuff" going on behind the scenes is really simple, and some of the stuff is incredibly complex. Seeing adversarial neural networks competing to effectively better each other and some of the results that come out from that... can be eye-opening!

ChatGPT and technologies of that ilk are a bit of a concern in some ways. Take my field of software engineering/development for example. We are already seeing people using AI tech to write code/script for them and it has set a somewhat dangerous precedent. Things like ChatGPT can return results with seemingly incredible levels of confidence, but it doesn't necessarily mean it is right or correct. Through laziness or inexperience, I have seen this myself when fellow developers have relied upon AI-lead script/code and simply used it without much thought (such is the perceived level of confidence in the results they've received from their queries). You simply cannot do that when the software you write can, for example, mean the difference between life/death for individuals or large proportions of the population! The more AI-derived code that comes into effect and goes unchecked, the higher the confidence in that software and the more likely future AI training evolution takes that as being "good" software/code. Interesting times.

Not ChatGPT, but a funny AI vid showing how AI learning can come up with unpredictable results and can find strange solutions to reach their end goal. Hold on to your papers. What a time to be alive!



I'm also reminded of another story (I can't find the video link) whereby researchers instructed an AI learning network to control a multi-legged spiderbot to move from one location to another, albeit with the minimum number of foot falls (i.e. minimum number of times the bot's feet touched the floor). After numerous rounds of training and leaving it alone to "learn", they were surprised to find it had made it from A to B with zero steps. Going back through the training and results data, they ran the simulation on the bot and found it had learnt to flip itself onto its back, and effectively shuffle on its back/knuckles - hence needing zero footsteps to get from A to B.

 

[Sir Nancy Flowers]

View attachment 1627296

Some of the "stuff" going on behind the scenes is really simple, and some of the stuff is incredibly complex. Seeing adversarial neural networks competing to effectively better each other and some of the results that come out from that... can be eye-opening!

ChatGPT and technologies of that ilk are a bit of a concern in some ways. Take my field of software engineering/development for example. We are already seeing people using AI tech to write code/script for them and it has set a somewhat dangerous precedent. Things like ChatGPT can return results with seemingly incredible levels of confidence, but it doesn't necessarily mean it is right or correct. Through laziness or inexperience, I have seen this myself when fellow developers have relied upon AI-lead script/code and simply used it without much thought (such is the perceived level of confidence in the results they've received from their queries). You simply cannot do that when the software you write can, for example, mean the difference between life/death for individuals or large proportions of the population! The more AI-derived code that comes into effect and goes unchecked, the higher the confidence in that software and the more likely future AI training evolution takes that as being "good" software/code. Interesting times.

Not ChatGPT, but a funny AI vid showing how AI learning can come up with unpredictable results and can find strange solutions to reach their end goal. Hold on to your papers. What a time to be alive!



I'm also reminded of another story (I can't find the video link) whereby researchers instructed an AI learning network to control a multi-legged spiderbot to move from one location to another, albeit with the minimum number of foot falls (i.e. minimum number of times the bot's feet touched the floor). After numerous rounds of training and leaving it alone to "learn", they were surprised to find it had made it from A to B with zero steps. Going back through the training and results data, they ran the simulation on the bot and found it had learnt to flip itself onto its back, and effectively shuffle on its back/knuckles - hence needing zero footsteps to get from A to B.

Honestly, what a reply

I has high hopes for the kind of reply that would come from you, expectations were exceeded.

I have no idea why but the final paragraph literally had me in stitches! Its a bit like the overly quoted phrase used by someone famous (Steve Jobs possibly?) about only hiring lazy people as they will complete a task with the minimal effort and therefore be more efficient.


brb, just pissing myself at the mental image of a robot shuffling itself along the floor on its back to 'beat' the system provided to it by its human overlords.

 

[SharkyUK]

brb, just pissing myself at the mental image of a robot shuffling itself along the floor on its back to 'beat' the system provided to it by its human overlords.

I'll try and find the video...

 

[Cookie]

View attachment 1627296

Some of the "stuff" going on behind the scenes is really simple, and some of the stuff is incredibly complex. Seeing adversarial neural networks competing to effectively better each other and some of the results that come out from that... can be eye-opening!

ChatGPT and technologies of that ilk are a bit of a concern in some ways. Take my field of software engineering/development for example. We are already seeing people using AI tech to write code/script for them and it has set a somewhat dangerous precedent. Things like ChatGPT can return results with seemingly incredible levels of confidence, but it doesn't necessarily mean it is right or correct. Through laziness or inexperience, I have seen this myself when fellow developers have relied upon AI-lead script/code and simply used it without much thought (such is the perceived level of confidence in the results they've received from their queries). You simply cannot do that when the software you write can, for example, mean the difference between life/death for individuals or large proportions of the population! The more AI-derived code that comes into effect and goes unchecked, the higher the confidence in that software and the more likely future AI training evolution takes that as being "good" software/code. Interesting times.

Not ChatGPT, but a funny AI vid showing how AI learning can come up with unpredictable results and can find strange solutions to reach their end goal. Hold on to your papers. What a time to be alive!



I'm also reminded of another story (I can't find the video link) whereby researchers instructed an AI learning network to control a multi-legged spiderbot to move from one location to another, albeit with the minimum number of foot falls (i.e. minimum number of times the bot's feet touched the floor). After numerous rounds of training and leaving it alone to "learn", they were surprised to find it had made it from A to B with zero steps. Going back through the training and results data, they ran the simulation on the bot and found it had learnt to flip itself onto its back, and effectively shuffle on its back/knuckles - hence needing zero footsteps to get from A to B.

I love the abuse of the physics system by the AI

 

[SharkyUK]

@Sir Nancy Flowers - I couldn't find the original video, but the first couple of minutes of this video show the result. I think it's hilarious! But nerd. 🤪

 

[Clio_fool]

It's all very intresting now but we are literally training our replacements. As Andy said programmers are just letting Ai do their job already and eventually the middle man (the programmer) gets cut out of the loop. How quickly that happens probably isn't even in our control, we probably can't turn it off now!

 

[SharkyUK]

It's all very intresting now but we are literally training our replacements. As Andy said programmers are just letting Ai do their job already and eventually the middle man (the programmer) gets cut out of the loop. How quickly that happens probably isn't even in our control, we probably can't turn it off now!

An AI produced this... honest... 🤪

 

[Clio_fool]

An AI produced this... honest... 🤪

View attachment 1627305

See the are already planning to take us! 🤣 Stay vigilant folks.

 

[Clio_fool]

OK I feel safer for now.

US Marines defeat Pentagon AI test by hiding in cardboard box

Proving that you don't need to worry about the robot takeover yet, a group of US Marines managed to defeat a Pentagon AI test by hiding in cardboard boxes

www.unilad.com

 

[McGherkin]

Bing's attempt at matching ChatGPT leaked the other day. Looks as impressive if not better than ChatGPT as it can search the web.

I think it should start at the right time but otherwise it's at 41 minutes 37 seconds onwards.


Really quite impressive how it can understand not just what was said but what the actual task was. First by understanding what the questioner wanted to know when asking what happened to Luke, then by actually calculating how many backpacks would fit in the back of a Tesla, then by analysing images on a website to decide whether two items would match, whether they had the same logo etc

 

[Clio_fool]

Bing wants to kill us all already and it's not even on general release!

Rogue AI chatbot declares love for user and says it wants to steal nuclear codes

Microsoft’s new AI chatbot went rogue during a chat with a reporter, professing its love for him and urging him to leave his wife.

www.lbc.co.uk

 

[botfch]

Bing's attempt at matching ChatGPT leaked the other day. Looks as impressive if not better than ChatGPT as it can search the web.

I think it should start at the right time but otherwise it's at 41 minutes 37 seconds onwards.


Really quite impressive how it can understand not just what was said but what the actual task was. First by understanding what the questioner wanted to know when asking what happened to Luke, then by actually calculating how many backpacks would fit in the back of a Tesla, then by analysing images on a website to decide whether two items would match, whether they had the same logo etc

Isn’t bing just chatgpt since Microsoft owns them both.

 

[Clio_fool]

Isn’t bing just chatgpt since Microsoft owns them both.

Yep.

 

[McGherkin]

I thought ChatGPT was google but the difference between ChatGPT and this is that the Bing version can search the internet live.

 

[Gally]

If you listen to the last couple if Vergecast (The Verge) you will get a handle on what the difference is and why the Bing version sounds a bit more mental.

I've used the basic version for a number of things in the last couple of weeks. Gives a nice outline and you can fill in the blanks if it's not something important.

 

[cjgower]

Have been using chatgpt for the initial drafting of email workflows and planning customer buying journeys at work. Also writing copy for e-shots.

Gives a general direction of email subjects and content, but I've found you have to really steer it in the right direction and be very specific to get it to have any real value.

Might as well learn to work with it.

 

[CrippsCorner]

If you listen to the last couple if Vergecast (The Verge) you will get a handle on what the difference is and why the Bing version sounds a bit more mental.

I've used the basic version for a number of things in the last couple of weeks. Gives a nice outline and you can fill in the blanks if it's not something important.

Read The Verge every day and had their podcast on earlier. Really is amazing stuff this Bing chatbot... apparently they're having to limit it to 5 replies now before it starts going off the rails lol.

 

[Hixle]

This s**t is f**king wild man.