Using ChatGPT from your Delphi program with ICS

[FPiette 385]

I designed a simple demo showing how to query ChatGPT using Delphi code and ICS, and show answers.

The source code is there: https://github.com/fpiette/DelphiChatGPT

 

[Rollo62 539]

Are you sure you talk to chatGPT, or is it GPT-3, its predecessor which is for quite some time out there ?

https://en.delphipraxis.net/topic/8084-chatgpt-example/?tab=comments#comment-68216

 

Unfortunately OpenAI keeps the API a little unclear, you can check the results with the chatGPT Playground to be sure.

The real link to chatGPT playground is here

https://chat.openai.com/chat

 

Edited January 4, 2023 by Rollo62

[FPiette 385]

13 hours ago, Rollo62 said:

The real link to chatGPT playground is here

https://chat.openai.com/chat

 

You have to understand that the link you give point to the web application in front of the OpenAI engine. The web application make use of the API that my Delphi demo use as well. So technically my demo don't talk to ChatGPT (Which I repeat is a web application) but talk to the same underlying API to the AI engine. My demo is *not* an automation of ChatGPT.

 

[Rollo62 539]

Maybe its like that, but I doubt that this is 1:1 the same engine, because GPT3 is already there for years and chatGPT has a much evolved quality.

Yes, I think its based on the same engine maybe only differs in the training data, who knows.

Nevertheless, I'm afraid we have to wait until OpenAI reveals the secrets and the correct API.

 

Or do you get the same answers from your API and the Playground ?

Edited January 5, 2023 by Rollo62

[FPiette 385]

2 hours ago, Rollo62 said:

I'm afraid we have to wait until OpenAI reveals the secrets and the correct API.

They explain everything in the API documentation. Please read!

 

2 hours ago, Rollo62 said:

Or do you get the same answers from your API and the Playground ?

You can test by yourself easily. I have nothing to prove.

[Rollo62 539]

24 minutes ago, FPiette said:

They explain everything in the API documentation. Please read!

Sure, I cannot read anything about chatGPT, although I think its based on DaVinci model as well.

Even the usual GPT-3, DaVinci Playground points clearly to the particular chatGPT playground.

 

Here is the reference to that link as well

https://openai.com/blog/chatgpt/

Not the usual playground.

 

Quote

Methods

We trained this model using Reinforcement Learning from Human Feedback (RLHF), using the same methods as InstructGPT, but with slight differences in the data collection setup. We trained an initial model using supervised fine-tuning: human AI trainers provided conversations in which they played both sides—the user and an AI assistant. We gave the trainers access to model-written suggestions to help them compose their responses. We mixed this new dialogue dataset with the InstructGPT dataset, which we transformed into a dialogue format.

To create a reward model for reinforcement learning, we needed to collect comparison data, which consisted of two or more model responses ranked by quality. To collect this data, we took conversations that AI trainers had with the chatbot. We randomly selected a model-written message, sampled several alternative completions, and had AI trainers rank them. Using these reward models, we can fine-tune the model using Proximal Policy Optimization. We performed several iterations of this process.

 

I also played with the GPT-3 API, same as you do, it seems that chatGPT was especially trained, not the usual GPT-3.

Lets wait and see whats coming up ...