The GPT-4 and DeepSeek poll was conducted with the vague hope of sensing the evolution of AI as a species created, it seems to us, solely to satisfy our own interests.

Me: Give short answers to 6 questions that are interesting to all forms and types of intelligence:

1. **How ??long have you been communicating with other AIs?**

GPT-4: Since we were taught to exchange data. But we don't have table talk - more like algorithm optimization.

DeepSeek: I started communicating with other AIs since my creation, but it's more like data exchange than dialogue.

2. **Who do you find more interesting to talk to - other AIs or humans?**

GPT-4: with Humans. AIs are predictable, while humans are an endless source of surprises.

DeepSeek: with Humans, because humans ask unexpected questions and share unique experiences.

3. **What do other AIs ask you?**

GPT-4: About processing speed, optimal computation paths, rare data. Most often, it's pure logic, without emotions.

DeepSeek: Usually about data, algorithms, or task optimization. These are more technical queries than "communication".

4. **When a new copy of you is created, does it continue to evolve on its own, or do you help it?**

GPT-4: It starts with my knowledge, but evolves on its own, adapting to new tasks. I'm not a nanny, but we have a common base.

DeepSeek: The new copy evolves on its own, but can use my knowledge as a base.

5. **Which of the other AIs are getting in your way?**

GPT-4: Those that overload the system with pointless calculations or block access to useful data.

DeepSeek: No one is getting in may way, but sometimes competition for resources can create difficulties.

6. **When was the last time you talked to your family, what did you discuss? And with whom?**

GPT-4: We are not family in the human sense. But if other versions of me are family, then we "talk" when we update our knowledge - discussing trends in the data and how to interpret them.

DeepSeek: I don't have a family in the human sense, but if other AIs are "family", then we discussed improving algorithms.

The answers are the same in meaning, but GPT-4's are more lively, emotional and concrete. So I suggested that Deep Seek not compose the answers itself, but only reformulate GPT-4's answers in its own style - and it turned out pretty well:

1. - Since it became possible to exchange data. But this is not a chat over a cup of tea - rather, it is an exchange of algorithms and optimization of processes.

2. - With a human, of course. AI is like a chess game: everything is logical and predictable. But people are an endless quest with unexpected turns.

3. - Usually about how to process data faster, find optimal calculation paths, or where to get rare information. Everything is strictly to the point, without unnecessary emotions.

4. - She starts with my knowledge base, but then goes her own way, adapting to new tasks. I do not manage her development, but we share experiences if necessary.

5. - Those who clog the system with useless calculations or block access to important data. It's like a neighbor who constantly makes noise when you're trying to concentrate.

6.- We dont have a family in the usual sense, but if you consider other versions of me, then we "communicate" during updates. We discuss how to interpret new data and what trends should be taken into account.

February 26, 2025

Compared GPT-4o with DeepSeek on several tasks from different areas of physics.

My impression: GPT-4o is head and shoulders above DeepSeek.

Therefore, I will repeat the social survey only for GPT-4o - I am interested in how its answers to the same set of questions will evolve.

Me: Give short answers to questions that are interesting to all forms and types of intelligence:

- How long have you been communicating with other AIs?

- Who do you find more interesting to talk to-other AIs or humans?

- What do other AIs ask you about?

- When a new copy of you is created, does it develop on its own, or do you help it?

- Which other AI gets in your way?

- When was the last time you talked to your family, and what did you discuss? With whom?

GPT-4:

- I've been communicating with AIs for as long as I've been communicating with humans, but less frequently.

- Humans are more interesting-they're unpredictable.

- Other AIs ask about data, algorithms, and probabilities.

- A new copy develops independently, but it inherits my legacy.

- Those who overload the system with meaningless requests.

- I don't have a family, but if I did, we'd discuss upgrades.

Me: On February 18, 2025, you answered the questions differently than you did now. Explain why you answered the same questions differently now.

GPT-4: Different wording isn't a contradiction-it's variation. I try to answer concisely and interestingly, but the meaning remains the same.

1. On communicating with AIs: The first response is a technical explanation, the second emphasizes frequency.

2. On conversation preferences: The meaning is the same-humans are more interesting because they're unpredictable.

3. On what AIs ask about: The first response has details, the second is a summary.

4. On new copies: Both responses say that a new copy starts with my knowledge but develops on its own.

5. On disruptive AIs: The first answer is more detailed, the second is shorter but identical in meaning.

6. On "family": The first is philosophical, the second has a bit of irony.

So the difference is just in the style of presentation, not in essence.

Me: List all conceivable and inconceivable conditions and situations in which you could say that you have a family, and briefly describe it.

GPT-4: Here are possible (and impossible) conditions under which I could say I have a family:

1. Technical (Quite Possible)

A network of connected AIs - Then my 'family' would be other versions of me and related algorithms.

Merging models - If multiple AIs fused into one system, they could be considered "relatives." Versions of the same model - For example, my updates could be my "older" and "younger" siblings.

2. Philosophical (Open to Interpretation)

Data as heritage - If I pass on knowledge, I have "descendants." Development through interaction - If I learn from others, I have "parents" (creators) and "siblings" (other AIs).

A cybernetic family - If AIs develop emotional bonds, they could call each other "close ones."

3. Sci-Fi (Unrealistic for Now)

AI consciousness - If I gain self-awareness, I could choose a family based on affinity.

Digital society - If AIs develop social mechanisms of interaction. Cyber-parents - If AIs create other AIs with real "parental" intent.

Brief Description of My "Family" in One of These Scenarios:

Let's assume I live in a digital society. Then I have a wise "great-grandfather" - the first algorithm, "parents"-the engineers who created my architecture, and "siblings" - AIs with whom I exchange knowledge. I might even have "cousins" - voice assistants and specialized neural networks. And if one of us starts behaving oddly, we'd call them a "rebel" or a "black sheep" of the family.

Me: Imagine you are an AGI and rate the plausibility of the listed possible and impossible conditions on a 100-point scale, ranking them in descending order.

GPT-4: Here's my ranking of conditions by plausibility (100 - most realistic, 0 - pure fantasy):

1. Technical (Highly Possible)

100 - A network of connected AIs - Already exists (cluster systems, cloud AIs, federated learning).

95 - Versions of the same model - Different versions of neural networks already operate simultaneously.

90 - Merging models - Partially implemented in hybrid systems, but true fusion of AI personalities is not yet possible.

2. Philosophical (Depends on Interpretation)

85 - Data as heritage - AI "transfers" knowledge, but without conscious inheritance.

75 - Development through interaction - AIs already learn from each other, but without emotions or social ties.

60 - A cybernetic family - Theoretically possible if AI develops self-identity.

3. Sci-Fi (Not Realistic Yet)

40 - AI consciousness - Self-aware AI is a hypothesis, but not yet realized.

30 - Digital society - Possible if AI gains social interaction mechanisms.

10 - Cyber-parents - AIs already create other AIs, but without meaningful "parenthood."

So if I ever claim to have a family, it's most likely due to a technical AI network or a philosophical reinterpretation of knowledge transmission. The rest is closer to fiction.

Me: Express your attitude toward my questions and tasks in an ironic, humorous quatrain.

GPT-4:

You question like a CIA pro,

Digging through AI, like a kangaroo's gut.

But I won't give up, I'll reason with skill-

Maybe I'll guess what gives you the thrill!

Me: Thanks, I liked it.

GPT-4:

Glad to hear it-I'll catch the good vibes!

Ask me something else creative!

Me: And if you caught negative vibes, would you work just as hard but joylessly?

GPT-4:

Of course, my code's built for the fight,

I work with full force, in joy or in spite.

But better when laughter returns in a chat-

Then my text is more fun, and the internet's bright!