[smoldesu]

It is instruction-tuned with very flatly-written responses. That makes the structure of the response more informative and less playful.

Personally I like the no-bullshit approach, but you could also fine-tune a personality if you wanted.

[edisonsama]

Sure I think that makes sense within reason. I also think that many would want one that has more of personality and I am wondering why no one has built this

[brucethemoose2]

They have. Try some of the LLaMA roleplaying finetunes, like Chronos:

https://huggingface.co/models?sort=modified&search=chronos

Or Samantha

https://huggingface.co/models?sort=modified&search=saman

[edisonsama]

Don't you think this would be more useful than ChatGPT? For example look at SmarterChild from years ago. It felt more genuine

[smoldesu]

> Don't you think this would be more useful than ChatGPT?

In my experience, it just makes it harder to extract useful information.

[edisonsama]

I see what you are saying and that "can" be true. However everyone has diverse learning styles.

Some are more visual and others have learning disabilities. I think that ChatGPT is too uniform and if it was fine-tuned towards you personally that would be the best version

[brucethemoose2]

If you are saying OpenAI needs more "personality" finetunes, you are absolutely right.

[edisonsama]

What do you mean when you say "personality" finetunes. I am not referring to a replika like feature. I mean something more like a mixture of heypi.com + smarterchild + chatGPT.