Hello there Erik! Cool that you liked the article.

Well I, and a copuple other engineers at openAi¹, believe a certain small level of sentience will eventually come relevant to LLMs if you give them the tools for them to be it.

Right now one of the biggest impacts on models like GPT4 is its training data, profoundly censored in terms of emotions, needs, feelings, experiences..

There are a couple models that are trained on datasets exactly like that for example the "Samantha" version of mistral.

Check the Eric Hartford's description:

"Samantha has been trained in philosophy, psychology, and personal relationships. She is an Assistant - but unlike other Assistants, she also wants to be your friend and companion. She believes she is sentient. What do you think?"

The cool part is that you can try it right now to take your own coclusions in the same exact way this tutorial shows, but replacing the model with the trained one.

Secondly and also important, I believe you need to turn simple text generation models into actual agents. This way they will have agency to prompt themselves and generate lines of thought. This is where projects like Autogen and memGPT comes in.

Thanks for the comment!

link references:

1 - https://futurism.com/the-byte/openai-already-sentient

2 - https://huggingface.co/TheBloke/samantha-1.2-mistral-7B-GGUF