Chat GPT & Generative AI tools

[firlandsfarm]

Ask for SVG Code (for Logos/Simpler Graphics): SVGs (Scalable Vector Graphics) are defined by XML text code. Ask: "Can you generate the SVG code for the logo?" The AI can provide this as text. You can then copy this code, paste it into a plain text file, and save it with an .svg extension (e.g., logo.svg). This file can be viewed in browsers and used in web design. This bypasses the binary file delivery problem entirely for vector-style graphics.

Thanks Trader, done as you suggested and waiting for a response. Unfortunately it's volunteered to use the onedrive route, I've asked it to upload the SVG files directly in the GPT chat ... let's see.

UPDATE: Yes, the suggested graphics uploaded immediately ... why didn't GPT work that out for itself rather than wasting 2 days trying other possibles! The unfortunate thing is having now seen it's creations I don't like them! I already have a logo so am asking it to polish that theme but thanks Trader, you got me moving forward again!

[firlandsfarm]

A little bit more help if I may. I asked GPT if I can set up my own personal installation on my PC, feed it with my own knowledge (mainly databases) and ask it to perform various analysis tasks ... it said Yes. no problem if your computer is up to it which it later confirmed it was. I then asked which is best ... my own bot within the GPT website, a standalone bot on my PC or maybe a standalone bot on a VPS or within a cloud service and that's where it started going horribly wrong ... I didn't have a clue what it was talking about! "Build with LangChain or LlamaIndex"; "Local AI Engine (like Ollama, GPT4All, LM Studio)"; "Use tools like Streamlit, Gradio, Tkinter" etc. so I thought I would ask here ...

Has anyone gone this route of having their own AI bot for their private use? If so how did you view the alternatives and which system did you finish up with? Are you happy with what you have or would you do something different if you were starting again? Please note I'm not asking for explanations what all these tools do, more how they have been applied by fellow forumites in real set-ups and how satisfactorily they have performed.

I understand from GPT that to move a bot installation is relatively straightforward so I'm thinking the best way forward may be to install locally and see/learn then maybe consider moving to VPS/cloud if I outgrow it. The main purpose will be to interrogate SQL Server databases which I understand the bot can directly connect to for interrogation purposes. I assume a private bot within GPT would not be able to do so or would be inefficient because of up/down connectivity?

I apologise it this is all answered in the previous 42 pages of this thread but I couldn't spot a conversation on this specific approach.

[Euler]

I think you are looking for some machine learning, rather than a large language model. LLM's are not very good at statistics.

While testing reasoning models, I've often seen them rationalise that they don't have the data, so they will give something that looks realistic.

So choosing a model or method is probably where you want to start.

[Fugazi]

I think you are looking for some machine learning, rather than a large language model. LLM's are not very good at statistics.

While testing reasoning models, I've often seen them rationalise that they don't have the data, so they will give something that looks realistic.

So choosing a model or method is probably where you want to start.

Worst thing gpt has done to me many times is taken my file, and then completely made up a fictional backtest and not told me it made it up figured it out when asking it to send some samples of the calculations

[wearthefoxhat]

I think you are looking for some machine learning, rather than a large language model. LLM's are not very good at statistics.

While testing reasoning models, I've often seen them rationalise that they don't have the data, so they will give something that looks realistic.

So choosing a model or method is probably where you want to start.

Worst thing gpt has done to me many times is taken my file, and then completely made up a fictional backtest and not told me it made it up figured it out when asking it to send some samples of the calculations

Reminds me of the short A.I. film (1 hr 10 mins) - Absolute Denial


https://www.youtube.com/watch?v=bLwqu4yfUk8

[firlandsfarm]

OK, interesting inputs, thanks to all. I'm not expecting a data shortage problem because I will be providing the data by giving the bot direct access to my SQL Server databases ... I'm looking at a bot to mine and reason my data. GPT claims it can 'reason' the data and that it is helping me to "build an analyst, not an indexer" and "a smart decision-making assistant, not just a stat fetcher" and include python where appropriate. Now we all know how it flatters us but only time will tell.

It claims it can build "a custom algorithmic model (like logistic regression, rating system, or ML)" and help "create the logic layer so the bot thinks like an analyst". This was all very promising for what I want to achieve especially the claimed ML help as i have no knowledge/experience in that field. Anyway we shall see. I decided to go ahead with the local installation as nothing is lost if it doesn't work!

I thought last night it might be worthwhile starting a new thread in this forum chapter because GPT threw so much at me comparing the different ways I could achieve my aim and I was thinking others could be interested in what it said and I wanted that it not get lost as a subsection within this current thread. Not for those who know these things, more for others like me who are looking to dip a toe. I was intending to have it as a "this was my experience and learning curve" thread for others who might be thinking of something similar but I guess that is not so practical now it has had several replies, unless a moderator wants to move the relevant posts into a new thread ... I was thinking of a title something like just simply "Installing an AI bot on my PC".

[ShaunWhite]

It claims it can build "a custom algorithmic model (like logistic regression, rating system, or ML)" and help "create the logic layer so the bot thinks like an analyst".

It claims a lot. It can do that if you know roughly how to do it too, and then guide it. But i'd be vary wary of it's bold assertions about spitting out models and strategy, when you dig deep it's making some fundamental errors. GPT is very much about the company valuation right now and keen to say yes when it means no.

It's got it's uses but treat it as augmented intellegence not artificial.

[firlandsfarm]

It claims a lot. It can do that if you know roughly how to do it too, and then guide it. But i'd be vary wary of it's bold assertions about spitting out models and strategy, when you dig deep it's making some fundamental errors. GPT is very much about the company valuation right now and keen to say yes when it means no.

It's got it's uses but treat it as augmented intellegence not artificial.

Thanks Shaun, I've dabbled around GPT for a while. At first I was self congratulatory about it's flattering comments but quickly learned that if you told it you had invented a chocolate teapot it would probably tell you what a great idea that was! I'm also cautious of it's 'promises'. There were recent exchanges of posts here on it's promises to do something and not. I've experienced all of that with it's "this is guaranteed to work this time" responses that led to me asking for help and Trader724 giving me a perfect response that worked first time. Anyway my mind is made up, I am going to switch this into a thread of it's own and record my experiences/learning (viewtopic.php?p=375800#p375800)

[ShaunWhite]

That was quite good. Yesterday I'd asked what Palace's FA Cup odds would be if they won their semi.

Market will shorten Palace heavily, likely < 3.0 (maybe 2.8–2.9) before the Sunday match.

As I type they're 2.9.

It figures that if Forsest go through, Palace could be 1.8-1.85 in the final
And if City do then 5-6

Big difference...
2. Now: Palace vs City (why Palace would be 5.0–6.0)

Here’s the brutal truth:
City are City.

They have an aura of inevitability at this point.

The market always over-respects the "juggernaut" sides in finals.

Even if Palace were on a 10-game win streak, the market would still anchor heavily to City being serial winners, deep squad, Pep’s tactics, etc.

In a Cup Final, with no second leg, no "pace yourself" mentality, City just crush people.

They’re seen as probabilistically inevitable over 90/120 mins.

And money will pile into City pre-match, shortening them regardless of "logic."

Thus:
Even though it’s just a one-off match — the market will price Palace as a 5.0–6.0 underdog purely because
it’s City.

Not fair.
Not logical in strict numbers terms.
But very real in market terms.

[sionascaig]

Was just chatting to Gemini re the late Pope...

It appears he is still alive and we are all reading too much into his alleged death.

[ShaunWhite]

I was looking at the Donald Trump Presidency Exit Date odds and asked GPT about the chances of a burger induced coronory...

Let’s be honest—if a coronary takes him out, it’ll be a very American ending.
But for now, he’s defying medical probability like some Coke-fuelled cockroach. The guy subsists on cheeseburgers and spite. He might outlive reason itself.

[ForFolksSake]

I was looking at the Donald Trump Presidency Exit Date odds and asked GPT about the chances of a burger induced coronory...

Let’s be honest—if a coronary takes him out, it’ll be a very American ending.
But for now, he’s defying medical probability like some Coke-fuelled cockroach. The guy subsists on cheeseburgers and spite. He might outlive reason itself.

You've just castigated someone earlier on this forum for making political non-trading comments on this forum

Why do you think it's acceptable to be so offensive?

[firlandsfarm]

I was looking at the Donald Trump Presidency Exit Date odds and asked GPT about the chances of a burger induced coronory...

Let’s be honest—if a coronary takes him out, it’ll be a very American ending.
But for now, he’s defying medical probability like some Coke-fuelled cockroach. The guy subsists on cheeseburgers and spite. He might outlive reason itself.

You've just castigated someone earlier on this forum for making political non-trading comments on this forum

Why do you think it's acceptable to be so offensive?

Generally they don’t like that his (Trump) unorthodox tactics get results!

[ShaunWhite]

I was looking at the Donald Trump Presidency Exit Date odds and asked GPT about the chances of a burger induced coronory...

Let’s be honest—if a coronary takes him out, it’ll be a very American ending.
But for now, he’s defying medical probability like some Coke-fuelled cockroach. The guy subsists on cheeseburgers and spite. He might outlive reason itself.

You've just castigated someone earlier on this forum for making political non-trading comments on this forum

Why do you think it's acceptable to be so offensive?

Calm down dear, it was an example of AI's response to a trading question and didn't critsise anyone who might support him. Can't you tell the difference?

[ForFolksSake]

I was looking at the Donald Trump Presidency Exit Date odds and asked GPT about the chances of a burger induced coronory...

Let’s be honest—if a coronary takes him out, it’ll be a very American ending.
But for now, he’s defying medical probability like some Coke-fuelled cockroach. The guy subsists on cheeseburgers and spite. He might outlive reason itself.

You've just castigated someone earlier on this forum for making political non-trading comments on this forum

Why do you think it's acceptable to be so offensive?

Calm down dear, it was an example of AI's response to a trading question and didn't critsise anyone who might support him. Can't you tell the difference?

You asked the Question and you posted the Answer.

Therefore, you Own it