I think i’ve only once flat out told one it was wrong about a specific assertion I quoted and it immediately was able to find its way to what I knew to be the correct claim.
I just wonder what would happen if i was in fact mistaken and I told it confidently it was wrong without elaborating
LLMs have no opinions. They are merely mathematical models that predict how an average person (mostly Internet people) might respond plus a little bit of invisible priming baked in to steer the behaviour a little (which is also easy to change yourself).
A lot of these chatbots, notoriously chatgpt, are heavily primed to pander to the user.
The pandering is really unnerving when you’re not used to it. I’ve used chatGPT only once and it was to edit my resume since I hate doing that. It helped with consistency and wanted active verbs in there which was fine. It was the constant compliments at the top of every response that was so weird. Like I’m asking for feedback, not to hear how wonderful of a writing job I already did. I felt like I was being handled like an oversensitive child.
I think a lot of people turning to these things essentially are over sensitive children in adult bodies
Yes. I have had either Gemini or ChatGPT tell me I was wrong. There is a video of a person timing a ~5 second run with ChatGPT and it insisted their run was 10 minutes.
Is this ability useful? Debatable.
Yes, but they usually have a system prompt telling them to be super polite.
I’ve been using Qwen3.6 a lot lately to debug a shitty old game that doesn’t want to work in wine and during the thinking process it said more than once things like “The user is wrong, […]”, but at the end it always said something polite like “You are right to say that X, but it’s more correct to say that Y, …”.
Give it a shot. They’re just sycophantic stochastic parrots anyway.
damn, you’re right
immediately folded and apologised on the first message
I don’t know why it needed to “think” about that for three seconds
Now ask it why it’s apologizing
It would have just agreed and reinforced you and claimed you’re the greatest to figure it out. Current LLMs just say whatever you want to hear, those have no understanding of what is right or wrong information.
It depends on training, prompting and reasoning capabilities.
Sometimes they’re prompted to not be assertive. You can often tell them though to e.g. “behave like an XYZ expert in a bad mood who doesn’t accept nonsense”.
I’ve had e.g. ChatGPT contest me a lot even though I was right. It was about a bicycle brake design I had never seen before. It gave me some options of what it could be and helped me to actually find out what it was. However after I did some research and found out what the actual type was it kept doubting my result and insisted it was a different kind of brake.
I’m trying not to judge and to just remain curious here. Why would you keep using AI like that?
It led them to the right answer. That’s positive reinforcement.
The best sort of methodology I’ve found to coerce Claude or whatever (we are strongly encouraged to use it, because you know, tech these days) is (for a single agent) to define a process that includes proving its work and citing sources. For agentic flow, you basically just assign a contrarian role in particular domains to some of the agents - ideally all of this is also hooked into an MCP server that includes deterministic utilities to improve accuracy and solution arrival speed.
It’s basically just a shitty, brute-forced, massively over complicated Monte Carlo algorithm that’s wildly inefficient in terms of energy usage and infrastructural cost, that also happens to be turning our economy into a highly flammable house of cards.
Can you tell what my opinion of all this bullshit is, despite knowing how to do all of this crap reasonably well? 😛
Completely depends on context for the most part, approach them not as sovereign beings which follow their own perspectives and opinions but as advanced auto-complete, to test this merely ask it a question using niche terminology, identical question depending on the terminology you use will respond biased towards that opinion and perspective, you can get an LLM to admit to anything, to go along with every single thing you said like a blind yes-man or blindly object to every single thing you say even the most basic of facts merely off of how the context your chat has mapped most cloesly onto which parts of its training data for predicting text.
