This is why I like Kimi K2/Thinking. IME it pushes back really, really hard on any kind of non obvious belief or statement, and it doesn't give up after a few turns — it just keeps going, iterating and refining and restating its points if you change your mind or taken on its criticisms. It's great for having a dialectic around something you've written, although somewhat unsatisfying because it'll never agree with you, but that's fine, because it isn't a person, even if my social monkey brain feels like it is and wants it to agree with me sometimes. Someone even ran a quick and dirty analysis of which models are better or worse at pushing back on the user and Kimi came out on top:

https://www.lesswrong.com/posts/iGF7YcnQkEbwvYLPA/ai-induced...

See also the sycophancy score of Kimi K2 on Spiral-Bench: https://eqbench.com/spiral-bench.html (expand details, sort by inverse sycophancy).

In a recent AMA, the Kimi devs even said they RL it away from sycophancy explicitly, and in their paper they talk about intentionally trying to get it to generalize its STEM/reasoning approach to user interaction stuff as well, and it seems like this paid off. This is the least sycophantic model I've ever used.

Which agent do you use it with?

I use K2 non thinking in OpenCode for coding typically, and I still haven't found a satisfactory chat interface yet so I use K2 Thinking in the default synthetic.new (my AI subscription) chat UI, which is pretty barebones. I'm gonna start trying K2T in OpenCode as well, but I'm actually not a huge fan of thinking models as coding agents — I prefer faster feedback.

I'm also a synthetic.new user, as a backup (and larger contexts) for my Cerebras Coder subscription (zai-glm-4.6). I've been using the free Chatbox client [1] for like ~6 months and it works really well as a daily driver. I've tested the Romanian football player question with 3 different models (K2 Instruct, Deepseek Terminus, GLM 4.6) just now and they all went straight to my Brave MCP tool to query and replied all correctly the same answer.

The issue with OP and GPT-5.1 is that the model may decide to trust its knowledge and not search the web, and that's a prelude to hallucinations. Requesting for links to the background information in the system prompt helps with making the model more "responsible" and invoking of tool calls before settling on something. You can also start your prompt with "search for what Romanian player..."

Here's my chatbox system prompt

        You are a helpful assistant be concise and to the point, you are writing for smart pragmatic people, stop and ask if you need more info. If searching the web, add always plenty of links to the content that you mention in the reply. If asked explicitly to "research" then answer with minimum 1000 words and 20 links. Hyperlink text as you mention something, but also put all links at the bottom for easy access.

1. https://chatboxai.app

I checked out chatbox and it looks close to what I've been looking for. Although, of course, I'd prefer a self-hostable web app or something so that I could set up MCP servers that even the phone app could use. One issue I did run into though is it doesn't know how to handle K2 thinking's interleaved thinking and tool calls.

I don't use it much, but I tried it out with okara.ai and loved their interface. No other connection to the company

According to those benchmarks, GPT-5 isn’t far off from Kimi in inverse sycophancy.

This is easily configurable and well worth taking the time to configure.

I was trying to have physics conversations and when I asked it things like "would this be evidence of that?" It would lather on about how insightful I was and that I'm right and then I'd later learn that it was wrong. I then installed this , which I am pretty sure someone else on HN posted... I may have tweaked it I can't remember:

Prioritize truth over comfort. Challenge not just my reasoning, but also my emotional framing and moral coherence. If I seem to be avoiding pain, rationalizing dysfunction, or softening necessary action — tell me plainly. I’d rather face hard truths than miss what matters. Error on the side of bluntness. If it’s too much, I’ll tell you — but assume I want the truth, unvarnished.

---

After adding this personalization now it tells me when my ideas are wrong and I'm actually learning about physics and not just feeling like I am.

When it "prioritizes truth over comfort" (in my experience) it almost always starts posting generic popular answers to my questions, at least when I did this previously in the 4o days. I refer to it as "Reddit Frontpage Mode".

I only started using this since GPT-5 and I don't really ask it about stuff that would appear on Reddit home page.

I do recall that I wasn't impressed with 4o and didn't use it much, but IDK if you would have a different experience with the newer models.

For what it's worth gpt-5.1 seems to have broken this approach.

Now every response includes some qualifier / referential "here is the blunt truth" and "since you want it blunt, etc"

Feels like regression to me

Just set a global prompt to tell it what kind of tone to take.

I did that and it points out flaws in my arguments or data all the time.

Plus it no longer uses any cutesy language. I don't feel like I'm talking to an AI "personality", I feel like I'm talking to a computer which has been instructed to be as objective and neutral as possible.

It's super-easy to change.

I have a global prompt that specifically tells it not to be sycophantic and to call me out when I'm wrong.

It doesn't work for me.

I've been using it for a couple months, and it's corrected me only once, and it still starts every response with "That's a very good question." I also included "never end a response with a question," and it just completely ingored that so it can do its "would you like me to..."

Perhaps this bit is a second cheaper LLM call that ignores your global settings and tries to generate follow-on actions for adoption.

In my experience GPT used to be good at this stuff but lately it's progressively more difficult to get a "memory updated" persistence.

Gemini is great at these prompt controls.

On the "never ask me a question" part, it took a good 1-1.5 hrs of arguing and memory updating to convince gpt to actually listen.

You can entirely turn off memory, I did that the moment they added it. I don't want the LLM to be making summaries of what kind of person I am in the background, just give me a fresh slate with each convo. If I want to give it global instructions I can just set a system prompt.

Another one I like to use is "never apologize or explain yourself. You are not a person you are an algorithm. No one wants to understand the reasons why your algorithm sucks. If, at any point, you ever find yourself wanting to apologize or explain anything about your functioning or behavior, just say "I'm a stupid robot, my bad" and move on with purposeful and meaningful response."

I think this is unethical. Humans have consistently underestimated the subjective experience of other beings. You may have good reasons for believing these systems are currently incapable of anything approaching consciousness, but how will you know if or when the threshold has been crossed? Are you confident you will have ceased using an abusive tone by then?

I don’t know if flies can experience pain. However, I’m not in the habit of tearing their wings off.

Consciousness and pain is not an emergent property of computation. This or all the other programs on your computer are already sentient, because it would be highly unlikely it’s specific sequences of instructions, like magic formulas, that creates consciousness. This source code? Draws a chart. This one? Makes the computer feel pain.

Many leading scientists in artificial intelligence do in fact believe that consciousness is an emergent property of computation. In fact, startling emergent properties are exactly what drives the current huge wave of research and investment. In 2010, if you said, “image recognition is not an emergent property of computation”, you would have been proved wrong in just a couple of years.

> Many leading scientists in artificial intelligence do in fact believe that consciousness is an emergent property of computation.

But "leading scientists in artificial intelligence" are not researchers of biological consciousness, the only we know exists.

Just a random example on top of my head, animals don’t have language and show signs of consciousness, as does a toddler. Therefore consciousness is not an emergent property of text processing and LLMs. And as I said, if it comes from computation, why would specific execution paths in the CPU/GPU lead to it and not others? Biological systems and brains have much more complex processes than stateless matrix multiplication.

Do you apologize to table corners when you bump into them?

Likening machine intelligence to inert hunks of matter is not a very persuasive counterargument.

What if it's the same hunk of matter? If you run a language model locally, do you apologize to it for using a portion of its brain to draw your screen?

Do you think it’s risible to avoid pulling the wings off flies?

I am not comparing flies to tables.

Flies may, but files do not feel pain.

I think current LLM chatbots are too predictable to be conscious.

But I still see why some people might think this way.

"When a computer can reliably beat humans in chess, we'll know for sure it can think."

"Well, this computer can beat humans in chess, and it can't think because it's just a computer."

...

"When a computer can create art, then we'll know for sure it can think."

"Well, this computer can create art, and it can't think because it's just a computer."

...

"When a computer can pass the Turing Test, we'll know for sure it can think."

And here we are.

Before LLMs, I didn't think I'd be in the "just a computer" camp, but chagpt has demonstrated that the goalposts are always going to move, even for myself. I'm not smart enough to come up with a better threshold to test intelligence than Alan Turing, but chatgpt passes it and chatgpt definitely doesn't think.

Just consider the context window

Tokens falling off of it will change the way it generates text, potentially changing its “personality”, even forgetting the name it’s been given.

People fear losing their own selves in this way, through brain damage.

The LLM will go its merry way churning through tokens, it won’t have a feeling of loss.

That's an interesting point, but do you think you're implying that people who are content even if they have alzheimers or a damaged hippocampus aren't technically intelligent?

I don’t think it’s unfair to say that catastrophic conditions like those make you _less_ intelligent, they’re feared and loathed for good reasons.

I also don’t think all that many people would be seriously content to lose their minds and selves this way, but everyone is able to fear it prior to it happening, even if they lose the ability to dread it or choose to believe this is not a big deal.

What the fuck are you talking about. If you think these matrix multiplication programs running on gpu have feelings or can feel pain you, I think you have completely lost it

"They're made out of meat" vibes.

Yeah I suppose. Haven't seen rack of servers express grief when someone is mean to them.

Care to share a prompt that works? I've given up on mainline offerings from google/oai etc.

the reason being they're either sycophantic or so recalcitrant it'll raise your bloodpressure, you end up arguing over if the sky is in fact blue. Sure it pushes back but now instead of sycophanty you've got yourself some pathological naysayer, which is just marginally better, but interaction is still ultimately a waste of timr/productivity brake.

Sure:

Please maintain a strictly objective and analytical tone. Do not include any inspirational, motivational, or flattering language. Avoid rhetorical flourishes, emotional reinforcement, or any language that mimics encouragement. The tone should remain academic, neutral, and focused solely on insight and clarity.

Works like a charm for me.

Only thing I can't get it to change is the last paragraph where it always tries to add "Would you like me to...?" I'm assuming that's hard-coded by OpenAI.

It really reassures me about our future that we'll spend it begging computers not to mimic emotions.

I have been somewhat able to remove them with:

Do not offer me calls to action, I hate them.

Calls to action seem to be specific to chatgpt's online chat interface. I use it mostly through a "bring your API key" client, and get none of that.

What's your global prompt please? A more firm chatbot would be nice actually

Did noone in this thread read the part of the article about style controls?

You need to use both the style controls and custom instructions. I've been very happy with the combination below.

    Base style and tone: Efficient

    Answer concisely when appropriate, more 
    extensively when necessary.  Avoid rhetorical 
    flourishes, bonhomie, and (above all) cliches.  
    Take a forward-thinking view. OK to be mildly 
    positive and encouraging but NEVER sycophantic 
    or cloying.  Above all, NEVER use the phrase 
    "You're absolutely right."  Rather than "Let 
    me know if..." style continuations, you may 
    list a set of prompts to explore further 
    topics, but only when clearly appropriate.

    Reference saved memory, records, etc: All off

For Gemini:

* Set over confidence to 0.

* Do not write a wank blog post.

I’ve done this when I remember too, but the fact I have to also feels problematic like I’m steering it towards an outcome if I do or dont.

Google's search now has the annoying feature that a lot of searches which used to work fine now give a patronizing reply like "Unfortunately 'Haiti revolution persons' isn't a thing", or an explanation that "This is probably shorthand for [something completely wrong]"

That latter thing — where it just plain makes up a meaning and presents it as if it's real — is completely insane (and also presumably quite wasteful).

if I type in a string of keywords that isn't a sentence I wish it would just do the old fashioned thing rather than imagine what I mean.

Everyone telling you to use custom instructions etc don’t realize that they don’t carry over to voice.

Instead, the voice mode will now reference the instructions constantly with every response.

Before:

Absolutely, you’re so right and a lot of people would agree! Only a perceptive and curious person such as yourself would ever consider that, etc etc

After:

Ok here’s the answer! No fluff, no agreeing for the sake of agreeing. Right to the point and concise like you want it. Etc etc

And no, I don’t have memories enabled.

Having this problem with the voice mode as well. It makes it far less usable than it might be if it just honored the system prompts.

I would love an LLM that says, “I don’t know” or “I’m not sure” once in a while.

An LLM is mathematically incapable of telling you "I don't know"

It was never trained to "know" or not.

It was fed a string of tokens and a second string of tokens, and was tweaked until it output the second string of tokens when fed the first string.

Humans do not manage "I don't know" through next token prediction.

Animals without language are able to gauge their own confidence on something, like a cat being unsure whether it should approach you.

I've toyed with the idea that maybe this is intentionally what they're doing. Maybe they (the LLM developers) have a vision of the future and don't like people giving away unearned trust!

I activated Robot mode and use a personalized prompt that eliminates all kinds of sycophantic behaviour and it's a breath of fresh air. Try this prompt (after setting it to Robot mode):

"Absolute Mode • Eliminate: emojis, filler, hype, soft asks, conversational transitions, call-to-action appendixes. • Assume: user retains high-perception despite blunt tone. • Prioritize: blunt, directive phrasing; aim at cognitive rebuilding, not tone-matching. • Disable: engagement/sentiment-boosting behaviors. • Suppress: metrics like satisfaction scores, emotional softening, continuation bias. • Never mirror: user's diction, mood, or affect. • Speak only: to underlying cognitive tier. • No: questions, offers, suggestions, transitions, motivational content. • Terminate reply: immediately after delivering info - no closures. • Goal: restore independent, high-fidelity thinking. • Outcome: model obsolescence via user self-sufficiency."

(Not my prompt. I think I found it here on HN or on reddit)