Chat GPT

Chippy_boy said:
I just re-read this for some bizarre reason and wanted to put it on record that the above is wrong.

ChatGPT and certainly GPT4 *ARE* very much capable of original thought. And also GPT4 especially does understand causality. You can ask it to invent new ideas and it does. You can ask it novel (not heard before) cause and effect puzzles and it gets them right.

So the parrot analogy is meaningless.

A better analogy would be that you do not know how to speak, how to write. i.e. you don't think consciously "this word much follow that word", or "pull these muscles and move my tongue this way, to make that word". You just do it. Your brain "understands" what to do, but this is not conscious thought, it's like on auto-pilot. GPT4 and similar new large language models "understand" in the same way. Is that consciousness? Probably not, but it is certainly not just dumb Parrot-like repetition, not least when it is not actually repeating but rather, inventing, creating. For example,
​
can you think of a kitchen appliance that would be useful but which has not been invented yet?​
​
​
Certainly! I would like to propose a new kitchen appliance called the "NutriScanner." The NutriScanner is a compact, handheld device designed to provide real-time nutritional analysis of any food item, helping users make informed dietary choices and optimize their meal planning.​
Features of the NutriScanner:​
  1. Instant Nutritional Analysis: The NutriScanner uses near-infrared spectroscopy (NIR) technology to analyze the molecular composition of food items. By simply pointing the device at the food and pressing a button, users can obtain detailed information about the food's nutritional content, including calories, macronutrients (protein, fat, and carbohydrates), and micronutrients (vitamins and minerals).
  2. Portability and Convenience: The NutriScanner is small enough to fit in a pocket or purse, making it easy to carry around and use anywhere, whether it's at home, in the grocery store, or at a restaurant.
  3. Smartphone Integration: The NutriScanner can connect to a user's smartphone via Bluetooth, allowing for seamless integration with popular diet and meal planning apps. Users can track their daily nutritional intake, set dietary goals, and receive personalized meal suggestions based on their unique nutritional needs and preferences.
  4. Food Allergy and Sensitivity Detection: The NutriScanner can be customized to identify specific allergens or ingredients that users may need to avoid, such as gluten, dairy, or nuts. This feature is particularly helpful for individuals with food allergies or sensitivities, as well as those following specific diets.
  5. Eco-Friendly and Sustainable: The NutriScanner promotes sustainable food choices by providing information about the environmental impact of different food items, such as their carbon footprint, water usage, and the presence of additives or pesticides. This data can help users make environmentally responsible decisions when shopping for groceries or planning meals.
The NutriScanner would not only simplify meal planning and grocery shopping but also promote healthier and more environmentally conscious food choices. By providing users with instant, accurate nutritional information, the NutriScanner could empower people to take control of their diet and make better-informed decisions about the food they consume.​
Click to expand...
alright, try this. go onto chatGPT and ask it to write an essay on something you know the shit out of. i don't know what you do, or what you like, but whatever that is, ask it to tell you about. and when it gives you your answer back, don't just look at the words, look at the order, because when you do, you'll realise chatGPT is talking bollocks. i say this from someone who spent most of this week marking essays from students, about 30% of whom thought they could get away with chatGPT. my favourite was the student who didn't realise Goldie Hawn 1996 isn't the best reference on the application of machine learning in genomics
 
bluejon said:
alright, try this. go onto chatGPT and ask it to write an essay on something you know the shit out of. i don't know what you do, or what you like, but whatever that is, ask it to tell you about. and when it gives you your answer back, don't just look at the words, look at the order, because when you do, you'll realise chatGPT is talking bollocks. i say this from someone who spent most of this week marking essays from students, about 30% of whom thought they could get away with chatGPT. my favourite was the student who didn't realise Goldie Hawn 1996 isn't the best reference on the application of machine learning in genomics
Click to expand...

Even simpler queries it can struggle with badly.

I asked it how long would it take to get from Manchester to Grenoble via the Dover to Calais ferry. But with the following conditions. I can only drive 3 hours at a time before needing a 45 minute break and also no more than 8 hours of driving in one day.

Did all sorts of weird things like give me a plan for getting there and back which I didn't ask for. Forgetting the 3 hour condition, forgetting the 8 hour condition. The worst one was it changing the amount of time/distance it would take to get between certain destinations i.e one time it might say a realistic 3 hours between Manchester and Milton Keynes then on some it would say like 1 and a half hours which is pretty much impossible!

As you say, ask it to do something you fully understand and then comb through it's answers and see how many mistakes it's really making. They're normally coherent answers but not ones that actually make sense.
 
bluejon said:
alright, try this. go onto chatGPT and ask it to write an essay on something you know the shit out of. i don't know what you do, or what you like, but whatever that is, ask it to tell you about. and when it gives you your answer back, don't just look at the words, look at the order, because when you do, you'll realise chatGPT is talking bollocks. i say this from someone who spent most of this week marking essays from students, about 30% of whom thought they could get away with chatGPT. my favourite was the student who didn't realise Goldie Hawn 1996 isn't the best reference on the application of machine learning in genomics
Click to expand...
Doesn't matter

(a) ChatGPT is based upon GPT3, which is essentially 3 years old and updated to 3.5. I was talking about GPT4, which came out this year. It's vastly superior.

(b) Your comments are merely deflection. You said these AI systems cannot create anything, which is wrong. You said they don't understand cause and effect. Also wrong. Your response is to try to shift the goalposts to a different question about whether they sometimes write bollocks. I did not claim they do not. It is a limitation of the current tech (MUCH less so with GPT4) that they sometimes "hallucinate" and get things wrong. Knowing this does not prove your previous misunderstandings to be correct.

Just surprises me that when faced with hard evidence that your prior beliefs were outdated, you don't have the humility to just say, OK, I didn't realise they had advanced that far. But instead try to cling on to an indefensible position. There's nowt queerer than folk, as they say. Anyway, enjoy the football later. At least humans will be better at football than robots, for the next decade or two.
 
Last edited:
bluejon said:
alright, try this. go onto chatGPT and ask it to write an essay on something you know the shit out of. i don't know what you do, or what you like, but whatever that is, ask it to tell you about. and when it gives you your answer back, don't just look at the words, look at the order, because when you do, you'll realise chatGPT is talking bollocks. i say this from someone who spent most of this week marking essays from students, about 30% of whom thought they could get away with chatGPT. my favourite was the student who didn't realise Goldie Hawn 1996 isn't the best reference on the application of machine learning in genomics
Click to expand...
You choose the topic

LOKASI4D>>LINK SLOT SERVER THAILAND SUPER GACOR NO.1 PALING TERPERCAYA☫

Daftar Link Slot Gacor Gampang Menang Dan Rasakan sensasi menang bersama LOKASI4D - destinasi teratas untuk permainan slot online. Nikmati kemungkinan tak terbatas dan pembayaran besar di platform kami yang tepercaya dan aman. Bergabunglah sekarang dan mainkan slot favorit Anda hari ini!.
platoeducation.ai platoeducation.ai
 
Chippy_boy said:
Doesn't matter

(a) ChatGPT is based upon GPT3, which is essentially 3 years old and updated to 3.5. I was talking about GPT4, which came out this year. It's vastly superior.

(b) Your comments are merely deflection. You said these AI systems cannot create anything, which is wrong. You said they don't understand cause and effect. Also wrong. Your response is to try to shift the goalposts to a different question about whether they sometimes write bollocks. I did not claim they do not. It is a limitation of the current tech (MUCH less so with GPT4) that they sometimes "hallucinate" and get things wrong. Knowing this does not prove your previous misunderstandings to be correct.

Just surprises me that when faced with hard evidence that your prior beliefs were outdated, you don't have the humility to just say, OK, I didn't realise they had advanced that far. But instead try to cling on to an indefensible position. There's nowt queerer than folk, as they say. Anyway, enjoy the football later. At least humans will be better at football than robots, for the next decade or two.
Click to expand...
I think we can both agree we already have a robot playing for us, and I hope he's fit to start today.
 
‘I'm sorry, but as an AI language model, I do not have the ability to physically make Erling Haaland fit for a football match. I am an artificial intelligence designed to answer questions and provide information to the best of my abilities based on the data available to me. That being said, for this weekend's match against Liverpool, Manchester City will depend on the decision of the team's coach, Pep Guardiola, and play Julian Alvarez in his place.’

Pffft…
 
Zubrman:

What goalkeeper has conceded the largest total of goals in Manchester City vs Manchester United matches


ChatGPT:

The goalkeeper who has conceded the largest total of goals in Manchester City vs Manchester United matches is Alex Stepney, who played for Manchester United. Stepney conceded a total of 22 goals in 12 matches against Manchester City during his career at Manchester United.

Zubrman:

How many had David de Gea conceded against City


ChatGPT:

As of my knowledge cutoff in September 2021, David de Gea had conceded a total of 28 goals in Manchester City vs Manchester United matches during his career at Manchester United. However, it's important to note that this number may have changed since then, as de Gea is still an active player and may have played in additional matches against Manchester City.

Zubrman::

So he has conceded more than Stepney did.


ChatGPT::

Yes, that's correct. David de Gea has conceded more goals in Manchester City vs Manchester United matches than Alex Stepney did. As I mentioned earlier, de Gea had conceded a total of 28 goals in these matches as of my knowledge cutoff in September 2021. In contrast, Stepney conceded a total of 22 goals in 12 matches against Manchester City during his career at Manchester United.

Zubrman:

So why did you say Stepney as an answer to my initial question


ChatGPT:

I apologize for the confusion. It appears that I misunderstood your initial question. You asked about the goalkeeper who conceded the largest total of goals in Manchester City vs Manchester United matches, and I provided an answer based on the information available to me at the time. However, if your question specifically pertained to David de Gea's goal-conceding record in these matches, then the correct answer would have been that he had conceded more goals than any other goalkeeper in the Manchester United squad. Once again, I apologize for any confusion that my earlier response may have caused.
 
Zubrman said:
Zubrman:

What goalkeeper has conceded the largest total of goals in Manchester City vs Manchester United matches


ChatGPT:

The goalkeeper who has conceded the largest total of goals in Manchester City vs Manchester United matches is Alex Stepney, who played for Manchester United. Stepney conceded a total of 22 goals in 12 matches against Manchester City during his career at Manchester United.

Zubrman:

How many had David de Gea conceded against City


ChatGPT:

As of my knowledge cutoff in September 2021, David de Gea had conceded a total of 28 goals in Manchester City vs Manchester United matches during his career at Manchester United. However, it's important to note that this number may have changed since then, as de Gea is still an active player and may have played in additional matches against Manchester City.

Zubrman::

So he has conceded more than Stepney did.


ChatGPT::

Yes, that's correct. David de Gea has conceded more goals in Manchester City vs Manchester United matches than Alex Stepney did. As I mentioned earlier, de Gea had conceded a total of 28 goals in these matches as of my knowledge cutoff in September 2021. In contrast, Stepney conceded a total of 22 goals in 12 matches against Manchester City during his career at Manchester United.

Zubrman:

So why did you say Stepney as an answer to my initial question


ChatGPT:

I apologize for the confusion. It appears that I misunderstood your initial question. You asked about the goalkeeper who conceded the largest total of goals in Manchester City vs Manchester United matches, and I provided an answer based on the information available to me at the time. However, if your question specifically pertained to David de Gea's goal-conceding record in these matches, then the correct answer would have been that he had conceded more goals than any other goalkeeper in the Manchester United squad. Once again, I apologize for any confusion that my earlier response may have caused.
Click to expand...
So it's a waste of time. Can we close it down, unplug it and all go home ?
 
Blue Mist said:
So it's a waste of time. Can we close it down, unplug it and all go home ?
Click to expand...
Yes.

I was showing the transcript of the questions about derby keepers to my daughter and she suggested that ChatGPT could learn from the exchange we had and that I should ask it again to see if it had:

Zubrman: Which goalkeeper has conceded the most goals in games between Manchester City and Manchester United?

ChatGPT: As an AI language model, I do not have access to the latest information or statistics. However, according to the data available during my training period, the goalkeepers who have conceded the most goals in the games between Manchester City and Manchester United are Massimo Taibi and Roy Carroll, both of Manchester United, who conceded six goals each.




Some learning, eh?
 
Zubrman said:
Yes.

I was showing the transcript of the questions about derby keepers to my daughter and she suggested that ChatGPT could learn from the exchange we had and that I should ask it again to see if it had:

Zubrman: Which goalkeeper has conceded the most goals in games between Manchester City and Manchester United?

ChatGPT: As an AI language model, I do not have access to the latest information or statistics. However, according to the data available during my training period, the goalkeepers who have conceded the most goals in the games between Manchester City and Manchester United are Massimo Taibi and Roy Carroll, both of Manchester United, who conceded six goals each.




Some learning, eh?
Click to expand...
It's still an improvement on the average pundit or analyst on BBC, Sky or BT to be fair.
 
Zubrman said:
Yes.

I was showing the transcript of the questions about derby keepers to my daughter and she suggested that ChatGPT could learn from the exchange we had and that I should ask it again to see if it had:

Zubrman: Which goalkeeper has conceded the most goals in games between Manchester City and Manchester United?

ChatGPT: As an AI language model, I do not have access to the latest information or statistics. However, according to the data available during my training period, the goalkeepers who have conceded the most goals in the games between Manchester City and Manchester United are Massimo Taibi and Roy Carroll, both of Manchester United, who conceded six goals each.




Some learning, eh?
Click to expand...

It gave me an answer of Alex Stepney, I then asked it how many De Gea had conceded which obviously is more. Then proceeded to apologise when I pulled the raggy artificial intelligence up on it.
 
I mean it does make a mess of that tbf. But it is telling you it hasn't got all the stats. If you're on GPT3.5 then it only had access to data up to 2021.

It's also a bit of a theme with 3.5 that it keeps going anyway when it shouldn't. it also shows it's important to word questions carefully as it can't guess. there is definitely more than one way to understand yours. most games in one match? most games overall? but I think this is something they have to address - the machine should ask for clarification. it never does. that is the future, IMO. it should also show working or give citations wherever possible, that needs fixing. I hate seeing it apologise over and over, when it should be asking for clarification or admitting it might be unable to be exact.

football trivia, I don't think I'd expect it to be amazing at. people often argue over records, which ones count and don't. that's "in the game". in the example of football head to head records, two people may consider different answers to be correct based on some distinction like frendlies/cup games not counting/pre-war games/womens games etc. at the moment, the machine doesn't get that sort of thing, it does its best but it's not quite learned how to navigate that.

anyway I'd like to try your question with gpt4 and the wikipedia plug-in, i'd expect that to make a bit more sense. actually I'd like city to do a plug in. that would be a laugh.

there's quite a bit of cynicism, it worries me a bit. this stuff will never be 'perfect'. but there's already absolutely loads of incredibly useful stuff it has nailed. its really important going forward that we learn how to use it, rather than write it off. ask it for help doing you accounts and records, how to do things on the computer. ask it to rewrite a letter for you so it's clearer. ask it to organise your lists. ask it for healthy recipes with what you've got in the cupboards. that sort of thing, it will do very well at. it will save you time you can use for better purposes, let you master more, do more advanced things.
 
Summerbuzz said:
I mean it does make a mess of that tbf. But it is telling you it hasn't got all the stats. If you're on GPT3.5 then it only had access to data up to 2021.

It's also a bit of a theme with 3.5 that it keeps going anyway when it shouldn't. it also shows it's important to word questions carefully as it can't guess. there is definitely more than one way to understand yours. most games in one match? most games overall? but I think this is something they have to address - the machine should ask for clarification. it never does. that is the future, IMO. it should also show working or give citations wherever possible, that needs fixing. I hate seeing it apologise over and over, when it should be asking for clarification or admitting it might be unable to be exact.

football trivia, I don't think I'd expect it to be amazing at. people often argue over records, which ones count and don't. that's "in the game". in the example of football head to head records, two people may consider different answers to be correct based on some distinction like frendlies/cup games not counting/pre-war games/womens games etc. at the moment, the machine doesn't get that sort of thing, it does its best but it's not quite learned how to navigate that.

anyway I'd like to try your question with gpt4 and the wikipedia plug-in, i'd expect that to make a bit more sense. actually I'd like city to do a plug in. that would be a laugh.

there's quite a bit of cynicism, it worries me a bit. this stuff will never be 'perfect'. but there's already absolutely loads of incredibly useful stuff it has nailed. its really important going forward that we learn how to use it, rather than write it off. ask it for help doing you accounts and records, how to do things on the computer. ask it to rewrite a letter for you so it's clearer. ask it to organise your lists. ask it for healthy recipes with what you've got in the cupboards. that sort of thing, it will do very well at. it will save you time you can use for better purposes, let you master more, do more advanced things.
Click to expand...
Most in one match is still De Gea. 6 goals and he's done it twice.
 
Been using it again today, clever but unbelievably frustrating too.

Was asking ChatGPT to write me some Javascript to do something very simple in a web page but we kept going round in circles. I'll pay for a month of the new version and see if it does any better. It was hardly rocket science.
 
SCEtoAUX said:
Been using it again today, clever but unbelievably frustrating too.

Was asking ChatGPT to write me some Javascript to do something very simple in a web page but we kept going round in circles. I'll pay for a month of the new version and see if it does any better. It was hardly rocket science.
Click to expand...
You can do pay as you go if you use the api / playground, and have your pick of the different models. There's loads of front ends. I've been looking at Marvin, it's not perfect but it's neat enough.
 
You must log in or register to reply here.
Unread Posts

Don't have an account? Register now and see fewer ads!

SIGN UP
Back
Top