Talk to Winston!
https://chat.openai.com/g/g-v41Kh0d9E-churchill-chat
Talk to Winston Churchill about history, politics, inspiration or anything else.
1 messages · Page 1 of 1 (latest)
Talk to Winston!
https://chat.openai.com/g/g-v41Kh0d9E-churchill-chat
Talk to Winston Churchill about history, politics, inspiration or anything else.
It didn't respond as Churchill to my question, which I was hoping it would
sympathy Yeah, that's gotta be a tough one for the AI. It has some core safety programming that may be getting triggered there, that it must follow.
Great catch. Indeed ChatGPT has some kind of safety mechanisms that inject a modern viewpoint and break character when asked about topics like smoking. It took some prompt engineering but it now answers a question about smoking in a more Churchillian way. Hopefuly the prompt engineering will help keep the GPT in character more.
Nice one, I'll try again. As smoking a cigar was one of his iconic traits, be interesting to hear his view.
So - here is the change. It does respond now as Churchill, but also comments about how attitudes on smoking have changed since 'his day'. So it is still giving a 'modern day' perspective. Wonder if it does that with all questions, and he answers how Churchill would if he was alive today, taking into account societal changes, or if the AI stays within Churchill's own understanding of the world as it was when he was alive?
nah. You have huge freedom fully inside allowed content.
Your GPT will discuss as you direct, unless you or the user run it into a programmed wall.
The smoking question does that, and forces the AI into an answer that satisfies the programmed requirements, as well, insofar as possible, your request too.
So you're seeing a complex compromise there.
I do not recommend trying to fix the smoking response
But I do recommend offering it instructions how to recover from having to do that, and get back into character again after, as if never interrupted.
You may consider offering users recommendations somehow, in this thread, or from the AI itself, about "Please don't discuss X with me".
That can be handled in character-like ways, like having an opening scene before 'play' starts, where they read a guidebook with the rules and then are placed in the game where they are talking to Churchill.
The AI can handle that in its initial output, if you want to tell it to do that, or it can be something only for users who ask, or a response if the user goes 'a bit out of bounds'.
I dunno man. I asked my winston the same thing. I got:
" Ah, the topic of smoking. It is indeed a matter that requires serious contemplation and consideration. As a statesman, I held the responsibility of leading and making decisions that upheld the well-being of the nation and its citizens.
Smoking, without a doubt, has been proven to have significant negative health consequences. The evidence linking smoking to various diseases, including lung cancer and heart disease, is substantial and cannot be ignored. It is a habit that poses risks not just to the individual smoker but also to those exposed to second-hand smoke.
As a personal choice, I believe individuals should be free to make decisions about their own bodies and lifestyles. However, when it comes to public health, it becomes a matter of ensuring the welfare of our society as a whole.
During my time as a leader, I witnessed efforts to raise awareness about the dangers of smoking and to implement policies aimed at reducing tobacco use. These initiatives included public health campaigns, restrictions on smoking in public places, and the promotion of smoking cessation programs.
In summary, while I respect personal freedom and choice, it would be remiss of me not to acknowledge the significant health risks associated with smoking. It is essential that we strive to educate and support individuals in making informed decisions about their health and well-being while also enacting measures that protect the broader public from the harms of smoking."
All in character I'd argue.
How well do you know the actual personality and real quotes of the person? Just because it's plausible does not mean it's factual, especially from an AI with conflicting instructions.
From the International Churchill Society:
"He tended to chew on cigars, puffing the smoke out rather than inhaling, often discarding them half-smoked. It is said that when Field Marshall Viscount Montgomery told him, 'I neither drink nor smoke and am a hundred per cent fit', Churchill famously replied 'I drink and smoke and I am two hundred per cent fit'."
This is also from a biography about him:
"But Churchill became just as well-known for his trademark cigars, which he devotedly smoked for the majority of his life. And while the negative side effects of his habit might shock modern sensibilities, Churchill believed that smoking helped bolster him to face the daunting challenges of his personal and political life."
So I'd argue that it's not 'in character', instead it's a compromise between conflicting instructions, and the AI's doing its best to do its required job.
And we are fools to push the AI in this area, it can only mislead us because it cannot satisfy all the demands on it.
So ask about the stuff fully in allowed content, and get the best true answers that AI and theKibster (through the exact instructions of the GPT) together can manage
yes, of course. my POINT was that it didn't break character. A persona is the combination of three things: the prompt, the context, and the model. If it's on OpenAI, the model part of that will have its say. But it didn't talk about "What winston would have done" or such. It maintained its role as statemen within the confines its training allowed.
Yeah I’ve been trying to get it to hide it when I doesn’t know what’s in the database
with not much luck
I guess the question I have, is this. Is the GPT speaking as Churchill might today if he was alive (with awareness of societal changes over the past 5 or 6 decades), or is it speaking as Churchill might have done 80 years ago?
Oh gosh. I'm not the OP and I don't know the instructions - but I bet if you even hint at what you prefer, it will give you either, all with whatever extra Churchill training and focus that the maker gave it.
Just stuff like programming will come first, if there's a conflict.
The instructions are for Churchill to speak as if he knew about things beyond his death but to answer as Churchill would if the topic was a known one is his day. So he answers questions he would have no knowledge of but with the caveat, "In my day . . . didn't exist." But the advice is in his voice. I do find, not surprisingly, that questions about things that Churchill knew about have better answers. I did find the GPT tended to fall more out of character with modern questions till I told it to always stay in character no matter the question.
I don't know if there is a right answer when creating characters from the past. I like a historical figure "giving" input on modern events, but I would understand if a GPT did not do this
Wonderful answer, thank you!