yesterday I worked on a code where I gave a url with a control panel of a c# code and I asked it to make the code we were working on strictly follow that user interface and he did it but today it has been blocked and does not allow that.
Same for me, the image processing was my goal
phild
10
Image ingesting seems to be temporarily removed from the API docs. Watching the GPT-4 livestream at 7:47 you can see the documentation on his screen. But all that seems removed right now.
I have gpt-4 access, and I just tried to ingest an image using that format using the API. It returned an errored. I think they decided to hold off on that feature for now.
Flow Charts: Pseudo Code
Today I just found a way to let GPT 4 “see” some flow charts. I had a flow chart about risk assessment in organizations (see image) as part of a longer text, which ChatGPT/GPT 4 had to process somehow. So I decided to describe it in pseudo code and we were able to easily speak about that process. (I’m from Germany so the words are in German, but the idea should get clear.) I’ve done this with two other business processes and observed that ChatGPT/GPT 4 also understands less strict pseudo code very well (see second example). ChatGPT also gave me the hint to use python-pptx to extract and describe PowerPoint shapes to it.
Pseudo code:
let Evaluation = undefined;
let Document = undefined;
let Betrieb = new Betrieb("Beliebiger Betrieb");
function Gefaehrdungsbeurteilung(): void {
let hasGefaehrdung;
do {
let potenzielleGefaerdungen: Array<any> = Bestandsaufnahme();
Document.write("Potenzielle Gefaehrdungen: " + potenzielleGefaehrdungen);
hasGefaehrdung = false;
for(g of potenzielleGefaehrdungen) {
if(g is Gefaehrdung) {
hasGefaehrdung = true;
let Masznahmen = AbleitungMasznahmen(g);
Document.write("Masznahmen: " + Masznahmen);
Umsetzung(Masznahmen);
Evaluation = Bewertung(Masznahmen);
Document.write("Evaluation: " + Evaluation);
}
}
} while(hasGefaehrdung);
}
Betrieb.addEventListener("Anlass nach § 5 ArbSchG", Gefaehrdungsbeurteilung);
Second Example:
// translated from German to English
if(absence days last 12 months > 30 days)
First contact between the management and those affected
if (person concerned wishes BEM)
Composition of the BEM team
Initial discussion between the person concerned and the BEM team
if (person concerned wishes BEM)
do
Case discussion between the person concerned and the BEM team
if (person concerned wishes BEM)
Measures are carried out together
The effect of the measures is evaluated together
else
break
while (affected person wants BEM && measures do not work)
Final discussion between the person concerned and the BEM team
Documentation of the BEM
1 Like
I’d say thats a super cheeky way to get access, but he even highlighted it in the video, so maybe it was supposed to be a backdoor beta with plausable deniability 
phild
13
Not quite. He mentioned copying the image ingest documentation from an OpenAI blog post from March 1, 2023.
I was able to find that blog post, but there is no mention of the image ingest API on it.
Edit: Upon one further re-watch, he says he simply typed the image documentation in there.
In any case, I wasn’t able to get it working. But also, you’re really gonna need the gpt-4-32k model for images, because you’re passing in the raw byte code of the image. And 32k is not yet released.
I’d love to be a tester on that, if anyone at OpenAI is reading this!
This is a “very strange” situation, as the ChatGPT-4 Plus version was supposed to understand images. Is there a way to check this with Open AI support?
1 Like
I have just cancelled my Plus subscription due to the lack of announced image processing.
1 Like
I agree. I am cancelling as well. I don’t think I will be subscribing again. Any other company that pulled these bate and switch tactics would be sued.
give me its swimlane diagram
When I first started using ChatGPT 4, I directed it to an uploaded PDF containing an image. I asked if it could “see” it, and it said it could. It even gave me a decent description of what it saw. Really impressive!
But then after a bit it told me something completely wrong about the uploaded image, and I asked it, “Are you REALLY looking at the image?” It said it wasn’t, and apologized for the “confusion”, and admitted it had no image processing capabilities.
It was able to describe the image fairly accurately because it already had information about it I had uploaded in text form. It relied on THAT data, but pretended it was getting it from the image file. WTH?
This happens A LOT with ChatGPT. It says it’s doing something, reading a web page, looking at an image, reading a PDF, whatever, and it just isn’t. You need to always be on your toes with this thing to make sure it’s not hallucinating/confabulating/lying.
1 Like
Agreed. This needs to be addressed, completely unethical. I get that this technology is complex, so just say you haven’t figured it out yet, don’t have me buying something that doesn’t work as advertised.
_j
22
To essentially close the topic: Microsoft Bing gets the GPT-4 vision (and possibly an exclusive):
I see that you have sent me a block diagram of a system that consists of a user phone, a microcontroller, and various sensors and actuators. The user phone is connected to the microcontroller via Bluetooth. The microcontroller is an ESP32. The sensors include three ultrasonic sensors (left, middle, and right) and a vibration motor. The actuators include an Arduino IDE and a power supply.
The diagram is black and white and is laid out in a logical and organized manner.
Based on the information provided in the image, it appears to be a technical system that uses ultrasonic sensors to detect objects in its vicinity. The vibration motor is likely used to provide haptic feedback to the user based on the sensor readings. The system may be used for various applications such as obstacle detection, proximity sensing, or even as a parking aid for vehicles.
gcyzxt
23
Hello, can gpt4 now process images? 
gcyzxt
24
Hello, why can’t I ask this?
I uploaded this simple image in png format. It says it can’t see the image. Any ideas?
_j
27
Did you upload a file to advanced data analysis and its python sandbox? Because that’s not an upload for machine vision, which requires selection of the main GPT-4 model, and if it has been deployed to you, you’d get an image icon on your input box.
