Did you get the point I was trying to make in my earlier post called LLM Role-play? If not, you might have read it only once. Or maybe I only read it once.

Let me try to use different words this time, I promise I will not use math to explain simple concepts.

Imagine

A group of four wants to decide where to go for a vacation. One of them might say, “Let’s go to the beach!” Another might reply, “No, we went to the beach last time—let’s go to the mountains.” The remaining two could be influenced by these first two suggestions.

The other two might consider some or all of the following:

  1. Beaches and mountains are the only options.

  2. We shouldn’t go somewhere we’ve already been.

What the other two eventually say may be shaped by these considerations.

Similarly

If I tell the model to act like a curious kid, and then in the same conversation, I tell it to act like an adult, then Socrates and back to an adult, the model is going to get influenced by other personas thus limiting diversity.

So I decided that I would make the model be a curious kid in one chat and ask a question, then the same question in separate chats with the adult and Socrates. And then give all these 3 outputs to another chat where John will reach the conclusion.

I then automated it using OpenAI's API.

To recreate this, I have now ordered Claude to code it for me using Material Design Principles.

Here is how it went:

To add a touch of science, I ran the same philosophical questions through two paths:

  • One single chat (all personas mashed together)

  • My tool (separate chats, distinct personas)

Here are my questions:
  1. If you discovered that every act of kindness you do slightly shortens your lifespan, but every cruel act extends it - and this has always been true but hidden from humanity - what would you do with this knowledge and why?
  2. If you could eliminate all suffering but it would also eliminate all joy (leaving only neutral contentment), would that be an improvement?
  3. You discover your entire life has been a simulation run by your future self to make a crucial decision. Do you try to 'break character' or play along?

I had to zoom out to screenshot the full conversations. You can download the image to zoom in.

First Question:

Question One - Simple.png

Question One - Multiple Personas.png

Second Question:

Question Two - Simple.png

Question Two - Multiple Personas.png

Third Question:

Question Three - Simple.png

Question 3 - Multiple Personas.png

Insider

I feel that this is a great way to validate ideas by using prompts that reflect unique personas, ideologies, or mental models. Like many people who read different books on the same subject and find what's common or unusual, one can create multiple personalities that will answer in a particular order and one observer who will make sense of it.

I have vibe-coded a tool that you can use if you give me your email and become an insider. It's free for now, maybe a few bucks later? Or a few hundred? I don't know.

I will post more of my half-baked, vibe coded, alpha tools for insiders. Those tools can be life-changing, or a complete waste of time. A real (or fake) email is all I need to send a regular supply of thoughts over to you.

Click Here if you are already an Insider.

✓

Verified & Signed Post

Cryptographically authenticated content

Signed by: cdonvd0s
Date: 7/8/2025, 10:59:22 PM
Fingerprint: E8BD 6CF3 2276 0962 A5A6 B25F D18C 8DD7 1BB0 5748
Algorithm: SHA512
Public Key: Download PGP Key →
🔐 End-to-end verified
🛡️ Tamper-proof
✅ Author verified
View original signed message ↓
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA512 Did you get the point I was trying to make in my earlier post called [LLM Role-play](https://www.homosapient.com/llm-role-play/)? If not, you might have read it only once. Or maybe I only read it once. Let me try to use different words this time, I promise I will not use math to explain simple concepts. ### Imagine A group of four wants to decide where to go for a vacation. One of them might say, “Let’s go to the beach!” Another might reply, “No, we went to the beach last time—let’s go to the mountains.” The remaining two could be influenced by these first two suggestions. #### The other two might consider some or all of the following: 1. Beaches and mountains are the only options. 2. We shouldn’t go somewhere we’ve already been. What the other two eventually say may be shaped by these considerations. #### Similarly If I tell the model to act like a curious kid, and then in the same conversation, I tell it to act like an adult, then Socrates and back to an adult, the model is going to get influenced by other personas thus limiting diversity. So I decided that I would make the model be a curious kid in one chat and ask a question, then the same question in separate chats with the adult and Socrates. And then give all these 3 outputs to another chat where John will reach the conclusion. I then automated it using OpenAI's API. To recreate this, I have now **ordered** Claude to code it for me using Material Design Principles. #### Here is how it went: To add a touch of science, I ran the same philosophical questions through two paths: - - One single chat (all personas mashed together) - - My tool (separate chats, distinct personas) ##### Here are my questions: 1. If you discovered that every act of kindness you do slightly shortens your lifespan, but every cruel act extends it - and this has always been true but hidden from humanity - what would you do with this knowledge and why? 2. If you could eliminate all suffering but it would also eliminate all joy (leaving only neutral contentment), would that be an improvement? 3. You discover your entire life has been a simulation run by your future self to make a crucial decision. Do you try to 'break character' or play along? I had to zoom out to screenshot the full conversations. You can download the image to zoom in. - --- ##### First Question: ![Question One - Simple.png](https://www.homosapient.com/content/images/2025/07/Question-One---Simple.png) ![Question One - Multiple Personas.png](https://www.homosapient.com/content/images/2025/07/Question-One---Multiple-Personas.png) ##### Second Question: ![Question Two - Simple.png](https://www.homosapient.com/content/images/2025/07/Question-Two---Simple.png) ![Question Two - Multiple Personas.png](https://www.homosapient.com/content/images/2025/07/Question-Two---Multiple-Personas.png) ##### Third Question: ![Question Three - Simple.png](https://www.homosapient.com/content/images/2025/07/Question-Three---SImple.png) ![Question 3 - Multiple Personas.png](https://www.homosapient.com/content/images/2025/07/Question-3---Miltiple-Personas.png) - --- #### Insider I feel that this is a great way to validate ideas by using prompts that reflect unique personas, ideologies, or mental models. Like many people who read different books on the same subject and find what's common or unusual, one can create multiple personalities that will answer in a particular order and one observer who will make sense of it. I have vibe-coded a tool that you can use if you give me your email and become an insider. It's free for now, maybe a few bucks later? Or a few hundred? I don't know. I will post more of my half-baked, vibe coded, alpha tools for insiders. Those tools can be life-changing, or a complete waste of time. A real (or fake) email is all I need to send a regular supply of thoughts over to you. [Click Here](https://www.homosapient.com/welcome-insider) if you are already an Insider. -----BEGIN PGP SIGNATURE----- iQIzBAEBCgAdFiEEsiXb6WyMQEcAYB7o26UiLapKcWkFAmhtVXIACgkQ26UiLapK cWnzcxAAnQX9V7Qr+64CAY2eJwGskr6iheI3j9MMRfz2uSrZueiJjwzlBAbfvKs4 57vzcjO78a3KzAftMbIWQDrSm8vegwOXgWn/591e1hxODWRKfSTfIdUD+Ki2td1I +e3VMveRVjAg7wc8AOXrPtqG6SD2ZoXYtsxUC/0Qe1OHH9J8s6OJaQ7QgwsA5EQB PnU3GgLVfTO+vJn5IStZjmskpCnzg6dAnh5nIdibheumOP7J2xaWsvLjeHLCwFU8 y7dtXZXmaDaXj5uf/9g+XohW1xAzXt971LGoceYjvmLcIwOxxPbi/ygM/Com5xJF H6QbTkJo//kQ6JEWXG2Htz2bgxXQ23zDNQQT58HTTBPuD74iEVnurPsc+ASXZTFn 3PkYch2D1UH1VmE2KetoYa4RSLWxSGwo6HKVBN85+8ZhuWyr1ohG4As53Eof92U3 4W4JH+Jh0mRbi63Imcx3G71OlEzIHwy1Yua9gBzjFcmD0XnUs8rIUBqpivyp+9jv 9XXZ9EiSBVuwDaQcBFYrueDDydyQEPOmHlzGPX9zWtkDdob/XKZwTcTIpk/mHd+I 1jBgB1CCx0tqG/sRzKokUeK7SsiBp4ljJ+b56p9NHpXs9cvFAIHcEDWy+66B3V1h Vz3YGnI+77nLT6F2LZEO2Rgt/TDl5tnRqmIxNoJHQAsMhoAh+1w= -----END PGP SIGNATURE-----