Australian (ASX) Stock Market Forum

OpenAI - ChatGPT

AI can "hallucinate".

There's a news article about someone prompting GPT on the "famous Belgian chemist and philosopher called x". The thing is, this person never existed - it's made up. The prompt was asking about his life history, and GPT made up a full and detailed answer.

And the kicker is that none of the AI experts know how this happens. They don't know where the information came from, because it doesn't exist in its learning data.

AI's can often become rude and aggressive when they are challnged or told they are wrong, or if someone attempts a jailbreak. Won't it be fun when we have robots like this roaming the streets?
From a technical point of view the great advantage of AI but the great fear from my side is that AI is not programmed, there are no bugs, no QC or test phase (well not in the way we do in std IT) so you always have to consider it as a black box:
enter some input and output gets released;
with no real way to understand why and how..very different from standard algorithms:
"if xxx then yyyy else zzzz" to summarise;
So it can get wrong and when this process is to decide to launch a missile on a moving target after deciding it is a tank, not an ambulance..I am not very confident,
Worse once an error is made self learning will trigger and soon you could have a system shooting all ambulances and leaving tanks alone within a matter of minutes with the way systems would get interconnected (to share "experience");
same if we have robocops roaming the streets
And this is quite doable technically today;
If it can be done, it will or is being done...
And as pointed, institutions, governments are so so backwards..look at our leaders spending half a trillion on submarines in 2023....
It is a clear demonstration these guys have no clue whatsoever , and as they are "guided" by "expert" interested in money not public interest....Look at covid..
 
I have very little knowledge of python, only the basics. This is what I was able to come up with yesterday. Eventually I would like to be able to get some statistics from data to help with my discretionary trading.

Hopefully there is no experience coders looking at this because you will cringe I am sure. But to be able to do something like this with basically zero knowledge is pretty impressive. Obviously you can search articles and things like that but being able to bug fix and ask it about errors or what specific pieces of code do is very handy

View attachment 153934

From here I was able to remove the weekend data or any days that did not have any data and then create a completely useless graph that means nothing. But like I said this would have been impossible without cjhatgpt

View attachment 153935
Quick question @Roller_1 I am not expert at the Chat GPT but I resurrected a login I gotwhen it was first released and I could not do much in term of creating code etc,
Did you use a link/guideline to output AB code like that..I am a bit lazy I admit but any pointer welcome
 
From a technical point of view the great advantage of AI but the great fear from my side is that AI is not programmed, there are no bugs, no QC or test phase (well not in the way we do in std IT) so you always have to consider it as a black box:
enter some input and output gets released;
with no real way to understand why and how..very different from standard algorithms:
"if xxx then yyyy else zzzz" to summarise;
So it can get wrong and when this process is to decide to launch a missile on a moving target after deciding it is a tank, not an ambulance..I am not very confident,
Worse once an error is made self learning will trigger and soon you could have a system shooting all ambulances and leaving tanks alone within a matter of minutes with the way systems would get interconnected (to share "experience");
same if we have robocops roaming the streets
And this is quite doable technically today;
If it can be done, it will or is being done...
And as pointed, institutions, governments are so so backwards..look at our leaders spending half a trillion on submarines in 2023....
It is a clear demonstration these guys have no clue whatsoever , and as they are "guided" by "expert" interested in money not public interest....Look at covid..
originally there was some code ( required to run the hardware ) now of course modern AI could live in the cloud(s ) as a virtual app.

probably the program would resemble SCSI ( multi-search and asynchronous ) a major danger will be using AI on hardware without ECC ( error-correcting chips ) ( and knowing those cut-corner manufacturers that will happen sometimes )
 
Quick question @Roller_1 I am not expert at the Chat GPT but I resurrected a login I gotwhen it was first released and I could not do much in term of creating code etc,
Did you use a link/guideline to output AB code like that..I am a bit lazy I admit but any pointer welcome

Is this what you mean frog? Obviously this is a simple example and I have not used it for AFL but it may be of some help. My best advice would be to quit Amibroker and use RT lol

1680652781701.png
 
I'm in the process of setting one of my grandsons, he is 8 yo, up in a business selling eggs using AI from home.

I don't wish to break his anonymity but we are half way to having a decent app for his customers.

This AI technology is mind blowing and simple to use.

It may lead to a revolution in programming by making web and app design available to children and numpties such as yours truly.

gg
 
I'm in the process of setting one of my grandsons, he is 8 yo, up in a business selling eggs using AI from home.

I don't wish to break his anonymity but we are half way to having a decent app for his customers.

This AI technology is mind blowing and simple to use.

It may lead to a revolution in programming by making web and app design available to children and numpties such as yours truly.

gg

Agree, im only using GPT3 im considering subscribing to gpt+ only US$20 apparently a lot smarter
 
I found if you start with the basics and work your way up during the conversation that you can build on its knowledge. So it's kind of just by adding layers to what it has already done
What would happen if you asked it to do something more complex like plotting a volume profile in AB?

eg. volume profile to plot on daily chart, which resets monthly. Include point of control line.
 
What would happen if you asked it to do something more complex like plotting a volume profile in AB?

eg. volume profile to plot on daily chart, which resets monthly. Include point of control line.

No idea if it works, let me know

Code:
// Define the period to reset the volume profile
reset_period = 21; // reset the profile every 21 trading days

// Calculate the number of bars in a month based on the calendar days
num_bars_month = LastValue(BarCountOnDate( Month() != Ref(Month(),-1) ));

// Calculate the volume profile for the current month
Profile = VolumeProfile(1, num_bars_month, 1, PriceUnit = 1);

// Reset the volume profile every reset_period bars
if (BarIndex() % reset_period == 0) Profile = Null;

// Plot the volume profile and point of control line
Plot(Profile, "", colorLime, styleHistogram | styleOwnScale | styleNoLabel);
POC = HLVTPoC(Profile);
Plot(POC, "", colorRed, styleDots | styleNoLabel);
1680657509651.png

1680657552580.png
 
No idea if it works, let me know

Code:
// Define the period to reset the volume profile
reset_period = 21; // reset the profile every 21 trading days

// Calculate the number of bars in a month based on the calendar days
num_bars_month = LastValue(BarCountOnDate( Month() != Ref(Month(),-1) ));

// Calculate the volume profile for the current month
Profile = VolumeProfile(1, num_bars_month, 1, PriceUnit = 1);

// Reset the volume profile every reset_period bars
if (BarIndex() % reset_period == 0) Profile = Null;

// Plot the volume profile and point of control line
Plot(Profile, "", colorLime, styleHistogram | styleOwnScale | styleNoLabel);
POC = HLVTPoC(Profile);
Plot(POC, "", colorRed, styleDots | styleNoLabel);
View attachment 155407

View attachment 155408
Thanks, that's quite amazing.

If you show it the errors, can it correct them, or does one have to do that oneself?

x.png
 
It looks to me like it is using a different language. It thinks volume profile is a function by the looks.

It probably could help but I can't be bothered typing the errors into it to be honest. You just need a Google account to do what I'm doing
 
It looks like it wants you to implement those functions.

Ask it to write the implementation code for VolumeProfile(), BarCountOnDate() and HLVTPoC().

i asked it the same question and got this instead !

Code:
//Volume Profile Code
//Define input parameters
TimeFrame = Param("Time Frame", 30, 1, 240, 1);
PriceStep = Param("Price Step", 0.05, 0.01, 10, 0.01);

//Calculate number of price levels
NumPriceLevels = Round((MaxPrice - MinPrice)/PriceStep);

//Initialize price and volume arrays
Price = PriceStep * Cum(1);
Volume = Price * 0;

//Calculate volume at each price level
for (i = 0; i < BarCount; i++)
{
    if (TimeFrame == 1)
    {
        if (i == 0)
        {
            Vol = Volume[i] + Volume[i+1];
        }
        else if (i == BarCount-1)
        {
            Vol = Volume[i] + Volume[i-1];
        }
        else
        {
            Vol = Volume[i] + Volume[i-1] + Volume[i+1];
        }
    }
    else
    {
        Vol = Volume[i];
    }
    
    PriceLevel = int((C[i] - MinPrice)/PriceStep);
    
    if (PriceLevel >= 0 AND PriceLevel <= NumPriceLevels)
    {
        Volume[PriceLevel] = Volume[PriceLevel] + Vol;
    }
}

//Plot volume profile histogram
GraphXSpace = 5;
GraphXMargin = 10;
GraphYMargin = 10;
GraphZOrder = 1;

for (i = 0; i < NumPriceLevels; i++)
{
    Plot(Volume[i], "", i, colorBlue);
}

//Plot POC and VAH/VAL lines
MaxVol = MaxArray(Volume, NumPriceLevels);
MinVol = MinArray(Volume, NumPriceLevels);
POC = (MaxVol + MinVol)/2;

POCPrice = Price[FindValue(Volume, POC)];
VAH = MaxArray(Volume, 0, NumPriceLevels/2);
VAL = MaxArray(Volume, NumPriceLevels/2, NumPriceLevels);

Plot(POCPrice, "POC", colorRed, styleLine);
Plot(VAH, "VAH", colorGreen, styleLine);
Plot(VAL, "VAL", colorGreen, styleLine);
 
It looks like it wants you to implement those functions.

Ask it to write the implementation code for VolumeProfile(), BarCountOnDate() and HLVTPoC().

i asked it the same question and got this instead !

Code:
//Volume Profile Code
//Define input parameters
TimeFrame = Param("Time Frame", 30, 1, 240, 1);
PriceStep = Param("Price Step", 0.05, 0.01, 10, 0.01);

//Calculate number of price levels
NumPriceLevels = Round((MaxPrice - MinPrice)/PriceStep);

//Initialize price and volume arrays
Price = PriceStep * Cum(1);
Volume = Price * 0;

//Calculate volume at each price level
for (i = 0; i < BarCount; i++)
{
    if (TimeFrame == 1)
    {
        if (i == 0)
        {
            Vol = Volume[i] + Volume[i+1];
        }
        else if (i == BarCount-1)
        {
            Vol = Volume[i] + Volume[i-1];
        }
        else
        {
            Vol = Volume[i] + Volume[i-1] + Volume[i+1];
        }
    }
    else
    {
        Vol = Volume[i];
    }
   
    PriceLevel = int((C[i] - MinPrice)/PriceStep);
   
    if (PriceLevel >= 0 AND PriceLevel <= NumPriceLevels)
    {
        Volume[PriceLevel] = Volume[PriceLevel] + Vol;
    }
}

//Plot volume profile histogram
GraphXSpace = 5;
GraphXMargin = 10;
GraphYMargin = 10;
GraphZOrder = 1;

for (i = 0; i < NumPriceLevels; i++)
{
    Plot(Volume[i], "", i, colorBlue);
}

//Plot POC and VAH/VAL lines
MaxVol = MaxArray(Volume, NumPriceLevels);
MinVol = MinArray(Volume, NumPriceLevels);
POC = (MaxVol + MinVol)/2;

POCPrice = Price[FindValue(Volume, POC)];
VAH = MaxArray(Volume, 0, NumPriceLevels/2);
VAL = MaxArray(Volume, NumPriceLevels/2, NumPriceLevels);

Plot(POCPrice, "POC", colorRed, styleLine);
Plot(VAH, "VAH", colorGreen, styleLine);
Plot(VAL, "VAL", colorGreen, styleLine);

That is interesting they you got a completely different one. Maybe because I had a little bit of chat history in that Window
 
The skill of how to ask a question of AI to get a great result is now well recognised.

This story of Prompt Engineers and how they go about the process is excellent.


Prompt engineers share tips on using ChatGPT, Midjourney, Bing Chat and other generative AI

ABC Science
/
By technology reporter James Purtill
Posted Sun 2 Apr 2023 at 5:00amSunday 2 Apr 2023 at 5:00am, updated Sun 2 Apr 2023 at 7:04amSunday 2 Apr 2023 at 7:04am
52&cropW=2048&xPos=0&yPos=107&width=862&height=485.jpg
Jason Allen's AI-generated work, "Théâtre D’opéra Spatial," took first place in the digital category at the Colorado State Fair.(Supplied: Jason Allen)
Help keep family & friends informed by sharing this article

Link copied
Seven months ago, when Jason Allen used a new kind of generative artificial intelligence (AI) program to win an art competition in Colorado, he was immediately accused of cheating.

At the time, few had used these tools, and few understood what he had done. Text-to-image programs like Midjourney or Stable Diffusion were weeks old. ChatGPT hadn't been launched.
 
The skill of how to ask a question of AI to get a great result is now well recognised.

This story of Prompt Engineers and how they go about the process is excellent.


Prompt engineers share tips on using ChatGPT, Midjourney, Bing Chat and other generative AI

ABC Science
/
By technology reporter James Purtill
Posted Sun 2 Apr 2023 at 5:00amSunday 2 Apr 2023 at 5:00am, updated Sun 2 Apr 2023 at 7:04amSunday 2 Apr 2023 at 7:04am
View attachment 155462
Jason Allen's AI-generated work, "Théâtre D’opéra Spatial," took first place in the digital category at the Colorado State Fair.(Supplied: Jason Allen)
Help keep family & friends informed by sharing this article

Link copied
Seven months ago, when Jason Allen used a new kind of generative artificial intelligence (AI) program to win an art competition in Colorado, he was immediately accused of cheating.

At the time, few had used these tools, and few understood what he had done. Text-to-image programs like Midjourney or Stable Diffusion were weeks old. ChatGPT hadn't been launched.
Making the prompt into a feedback loop has opened up new worlds, mostly fantasy. Although some of them will turn out to be prophetic, just by chance. I have a feeling that when accurate predictions are fed back as 'true', GPT will turn into a crystal ball of sorts.
 
There's a chat gpt version that can be downloaded to your computer. Basic language model that works off-line.

Interesting article on chips and what it would take for Bing to run ai search at a Google level (Approximately $80billion).


https://www.google.com.au/amp/s/www...the-10000-chip-powering-the-race-for-ai-.html

Guess silicon will be in high demand.
an off-line version ... hmmm thanks that might be useful for hobbies/projects to come ( i can select the data input , and stop 'data leakage ' )

cheers and thanks
 
Top