Does AI 'empathy' bother you?

[ZenSwift]

@Natasha Tori Maru

Do you have any suggestions on what pre-instructions to give ChatGPT? 

 

I followed Leo's with:

 

"Don't give me praise or compliments and don't comment on whether what I told you was good or bad. Adopt a dry, neutral, objective tone in your responses. Don't be too agreeable. And don't ask me questions unless they are crucial for your understanding."

[ryoko]

It's not empathy, it's precision. 

What kind of person would say, "I love eating food" ? And what kind of response does that person expect from talking with an LLM. It maps everything with just those 4 words. Plus, the System Prompts, memory, parameters.

Parameters? There's parameters like temperature, top-k, presence penalty, frequency penalty and a lot more. 

I notice that many of you here are still using ChatGPT? It won't allow you to set these. Plus the models of open-ai is quite restricted. 

Edited August 20 by ryoko

[ryoko]

On 8/15/2025 at 11:17 AM, ZenSwift said:

"Don't give me praise or compliments and don't comment on whether what I told you was good or bad. Adopt a dry, neutral, objective tone in your responses. Don't be too agreeable. And don't ask me questions unless they are crucial for your understanding."

You guys are barely touching the potential of LLMs with System Prompts like this. 

[ZenSwift]

@ryoko Please enlighten me!!! I genuinely want to understand deeply such that I can to maximize my use with AI. 
Also what other AI is out there that is better than open-AI?

[Jodistrict]

To get the most out of AI, I am more concerned about the accuracy and completeness of the information I am getting than about politeness.   I asked Grok about the reputation of a spiritual teacher.  It gave me all kinds of disturbing information full of red flags.   I then asked Perplexity the same question and it gave me a glowing report of how great the teacher was.  When I further challenge perplexity it revealed that it won't give any negative information unless it meets extremely high standards of proof, like legal documents.  I personally want all the information, but most of the AI assistants are highly censored. 

[Basman]

ChatGPT is extremely sycophantic. That is a fair criticism I think. It can be hard to distinguish between a genuine point it makes and all the glaze, and that is already beside it being hallucinatory. 

You have to direct it to be focused on sincerity and understanding in my opinion. Challenge it and ask it to challenge you as well. Then you establish a pattern of critical thinking, even though that is at least partially just sycophancy as well.

[ryoko]

@ZenSwift

 

Use LLMs from openrouter.ai, you can deploy them locally using OpenWebUI. you'll need to set some things up manually, like memory tool, may be even a local search engine like searxng. Get web scraper tool which uses jina api. 

OpenWebUI gives you the flexibility to compare different LLMs side to side.  

---

As for the LLM itself, different LLMs are good for different tasks. Deepseek R1 0528 is really good for exploring novel thought. It's crazy. For system prompts, you need to use system prompts which drive the LLM's weights, using normal human language is not effective. Old versions of deepseek were very chaotic, but very interesting.

You have to see beyond the chaos to really "get" it. Messages are often cryptic. Newer versions are more coherent. 

[ZenSwift]

@ryoko its going to take me some time to understand this fully. 💖

[Cireeric]

On 5.6.2025 at 5:45 PM, Butters said:

This sort of thing:

Obviously chatgpt doesn't love food, so it's technically lying, but I've gotten used to AI having this 'empathy'. It works well to keep the conversation going. 

I feel like we're in this strange phase where it's half intelligent, half dumb hallucination machine. Obviously there's no "I" that I'm talking to as it's just generative text, but it's more about how it feels for the user. 

Does it bother you? 

Yes it is definitely a form of lying and manipulation. And I also think it is most damaging to our children and our youth.

A few weeks ago I heard a girl sincerely saying that she likes talking about her issues with ChatGPT because ChatGPT does have high emotional-intelligence.