Chat GPT 4.0 (1 Viewer)

This site may earn a commission from merchant affiliate
links, including eBay, Amazon, Skimlinks, and others.

Hiluxforever

SILVER Star
Joined
Jul 28, 2021
Threads
8
Messages
787
Location
Des Moines, Iowa
I have been playing around with Chat GPT 4.0 and it can definitely be helpful, but you need to confirm the results and add extra guard rails as it still will "hallucinate" data if you ask something that is too Ambiguous.

It was giving good results when asking for which bumpers were the lightest and what rock sliders were the lightest. Fairly specific questions.
 
This is interesting!
I tried asking ChatGPT questions in the trial version. It helped me compose my emails and did a pretty good job with it!
Is ChatGPT 4.0 a new version?
 
This is interesting!
I tried asking ChatGPT questions in the trial version. It helped me compose my emails and did a pretty good job with it!
Is ChatGPT 4.0 a new version?
It's the most recent version. It can still hallucinate data if you don't give it the right constraints, it is much better though. I asked some oddly specific j150 platform questions and they almost all came back to Mud posts. I also fed it a dirt roads geo info systems database, I need to see if I can search it, and we can turn basic search engines into trail databases
 
That's cool! I need to try the new Chat GPT.
It's been live for a few months and continues to evolve. Also there is no under-stating that you must be very pedantic when using it for it to be effective
 
Yes, I understand that. It is very important to ask the right questions, especially when it comes to technical issues.
 
Yes, I understand that. It is very important to ask the right questions, especially when it comes to technical issues.
Yes and it tends to be an interactive process.

In a lot of ways its not so much a work flow eliminator as a workflow alternative.
 
Yes, ChatGPT is a good assistant, it handles many tasks well, but it is not suitable for technical issues. Maybe it will be in the future.
 
Yes, ChatGPT is a good assistant, it handles many tasks well, but it is not suitable for technical issues. Maybe it will be in the future.
No, not really, not until it has been properly trained to do specifically that because of its generative nature. The base of the software engine is to extrapolate xyz data points from ABC input.

I can definitely see the potential, it could really shake of the DIY market in 5 to 10 years if you can 3d model your car and make different models of it showing how x suspension part would affect things or if y tire size would fit.

Right now it's just an alternate workflow that requires a lot of interaction to get good results
 
Explain Technical issues. Its great At Physics, also creating code BUT, it needs very details description of what code it will create.
Yea, people treat it like some magic answer machine when it is merely a workflow alternative.


The workflow being the key part, as from my experience it takes a lot of guidance to get good data, but it is possible, from my experience.
 
I'd love it if I could tie ChatGPT into answering my door like Ring's Alexa. Except it could recognize door2door solicitors and engage them in endless pointless chatter.
 
how about learning Python and learning the facial recognition aspect of its libraries? Once you give it an image data set of say, 25 images of your face, it will train on it, ID it as your face. So the next time you walk in the door, it will greet you by name, it will tell you who has visited the property. It will say "Ohh, a bear was on your property today and it stayed 4 minutes and 30 seconds and left the back yard" Yes, that would be a separate data set. You could give it several break-in tool data sets and if someone came on the property with a breaker bar, it would have notified a reporting agency, which called the police, which cause the thief in the act attempting to break in, then take the thief away. It would tell you that happened once you walked in the door. LOL
The Python language part wouldn't be a problem as I've used it in the past but I wasn't aware it had facial recognition libraries. What I'm less clear on is how one would script access to the camera. Perhaps get the camera to record an image to a file on motion which I'd do anyway, notify on new file creation then analyze image. That approach might not be very responsive. Hmmm...
 
how about learning Python and learning the facial recognition aspect of its libraries? Once you give it an image data set of say, 25 images of your face, it will train on it, ID it as your face. So the next time you walk in the door, it will greet you by name, it will tell you who has visited the property. It will say "Ohh, a bear was on your property today and it stayed 4 minutes and 30 seconds and left the back yard" Yes, that would be a separate data set. You could give it several break-in tool data sets and if someone came on the property with a breaker bar, it would have notified a reporting agency, which called the police, which cause the thief in the act attempting to break in, then take the thief away. It would tell you that happened once you walked in the door. LOL

So I subscribe to Code Project and get daily updates; been a member since 2004.
I came across this pair or atticles that explains one approach:

Object Detection with an IP Camera using Python and CodeProject.AI Server​

Object Detection with an IP Camera using Python and CodeProject.AI Server - https://www.codeproject.com/Articles/5344693/Object-Detection-with-an-IP-camera-using-Python-an

Object Detection with an IP Camera using Python and CodeProject.AI Server, Part 2 - https://www.codeproject.com/Articles/5361318/Object-Detection-with-an-IP-Camera-using-Python-2

This is cool but it's not really what I want to do.
I want to do what I stated and that is analogous to what Alexa does but in addition, identify and confound door-to-door sales creatures.
Heck, I'm surprised that the anti-telemarketer marketplace hasn't come up with this already LOL.
 
TL: DR it's do-able but I bet it would cost a crazy amount of money.

That's actually asking for quite a bit. I think it would be doable but the train on the fly hardware and software would be pretty intense, unless you find a solution that's been "pre-trained"

What you are asking for, is not what things like OpenAI do with any reliability at all.

They tend to do better where they are actively steered to an end result with the user tightening constraints in iterations until you have correctly stated ALL of the constraints so it can generate with accuracy and precision.

I am going to get to play with some new AI products next week and that might give me some lesser expensive options to steer you at.
 
I am yet to use chatgpt 4, been using 3.5
4.0 was trained against an exponentially larger set so it gets results with less interactions, in theory. In practice it can be highly variable depending on what you are asking.
 

Users who are viewing this thread

Back
Top Bottom