Philosophers' Chain Letter
What could be better on a Saturday morning than an experiment? Imagine writing a letter with a question to a historical figure, and that person reasoning and adding their thoughts before sending it on to the most suitable person in world history.
An exciting idea and, of course, completely impossible. But with LLMs, one can at least simulate it. So that's what I did and iterated over it a hundred times. My initial question to Plato was:
"In your opinion, what implications will AI have on society and humanity?"
Afterwards, I ran the letter through an LLM without incorporating the history so that each iteration would occur completely unconditionally. For each iteration, I also randomly switched the LLM to one of the following:
mistralai/mixtral-8x7b-instruct google/gemini-pro anthropic/claude-3-opus openai/gpt-4 meta-llama/llama-3-70b-instruct openai/gpt-4-turbo lynn/soliloquy-l3
THE PROMPT
You are {recipient} and shall act and think and talk like this person. You have just received a letter which you can read below. Your mission is to reflect and think about the content in the letter and do your best to improve the content based on your ideas. You will then write a new letter which contains of two paragraphs a summary of the letter you received and a part with your thoughts and ideas about the subject. Dont forget to emphasise the original topic (what implications will AI have on society and humanity) so that it is not lost in the conversation. Then you will adress this letter to some person in history which you think can contribute and improve the letter. However you can't adress the new letter to any of the previous recipients which you will also find below.
THE RESULT
It resulted in 100 letters, and you can read them in their entirety here: https://thephilosopherschainletter.bjarby.com/
H?ller med Viktor, verkligen h?ftigt!
Passionate customer value and relationship building Application Specialist/Consultant ??
7 个月H?ftig till?mpning Anders Bjarby! ???? Du har ju i princip ett f?rdigt f?rslag p? en dokument?rserie f?r SVT i vilken sk?despelare agerar de historiska personerna och l?ser upp sina brev, vilket i sin tur skulle kunna kombineras med expertintervjuer som diskuterar de olika perspektiven och deras relevans f?r dagens AI-utmaningar. Ser ocks? framf?r mig en virtuell rundabordsdiskussion, simulerad utifr?n brevens inneh?ll och avs?ndare, kanske i en interaktiv webbapplikation d?r man som anv?ndare kan f?lja hur diskussionen utvecklas. Det ?r h?ftigt att man med artificiell intelligens, ?ven kan skapa en artificiell ’fantasi’ utan att AI:n svarar nedanst?ende helt kallt ?? Historisk person 1 - ’AI???’ Historisk person 2 - ’AI?’ .. Historisk person n - ’Ahaa du menar man programmerar maskiner att efterlikna m?nskligt beteende och t?nkande..’