[awithrow]

It feels like I'm fighting uphill battle when it comes to bouncing ideas off of a model. I'll set things up in the context with instructions similar to. "Help me refine my ideas, challenge, push back, and don't just be agreeable." It works for a bit but eventually the conversation creeps back into complacency and syncophancy. I'll check it too by asking "are you just placating me?" the funny thing is that often it'll admit that, yes, it wasn't being very critical, and then procede to over correct and become a complete contrarian. and not in a way that's useful either. very frustrating. I've found that Opus 4.6 is worse about this than 4.5. 4.5 does a better job IMO of following instructions and not drifting into the mode where it acts like everything i say is a grand revelation from up high.

[post-it]

> I'll check it too by asking "are you just placating me?" the funny thing is that often it'll admit that, yes, it wasn't being very critical, and then procede to over correct and become a complete contrarian. and not in a way that's useful either.

It's not admitting anything. Your question diverts it down a path where it acts the part of a former sycophant who is now being critical, because that question is now upstream of its current state.

Never make the mistake of asking an LLM about its intentions. It doesn't have any intentions, but your question will alter its behaviour.

[godelski]

  > Your question diverts it down a path where it acts the part of a former sycophant who is now being critical

I think people really have a hard time understanding a sycophant can be contrarian. But a yesman can say yes by saying no

https://news.ycombinator.com/item?id=47484664

[cheesecompiler]

> a yesman can say yes by saying no

What a great way to summarize LLM behaviour in 2026

[noisy_boy]

>> a yesman can say yes by saying no

>What a great way to summarize LLM behaviour in 2026

Well they have been trained on words spoken by humans and that has been a human behaviour since time immemorial. E.g.: "I do not agree with you that you were wrong. I do apologize for my strong disagreement but we actually do need your continued guidance desperately."

[Terr_]

> It doesn't have any intentions

Yeah, and in a way it's even worse than that, since there's another layer of cognitive illusion: "It" doesn't exist.

The LLM algorithm is an ego-less document-generator, often applied to growing a document that resembles dialogue between two fictional characters.

So when your human-user character is "asking" the AI assistant character to explain its intentions, that's the same as asking a Count Dracula character to describe what it "really feels like" to become a cloud of bats.

You'll see something interesting, but it'll be what fits trained story-patterns rather than what any mind introspects or perceives.

[mvkel]

> it'll admit that, yes, it wasn't being very critical, and then procede to over correct and become a complete contrarian

Which is also placating you

[awithrow]

Yes exactly! It's very frustrating. My guess is it drives engagement. But some model that tells me what I want to hear all the time is useless.

[cruffle_duffle]

An alternate way of thinking about it is LLMs have no reflection capability. Literally any “reflection” it claims to have about its decision making is made up. It has absolutely no way know that what it said was based on some ancient proverb, the phase of the moon or cold hard rational thought.

[layer8]

I think “admit” here is just a description of what the LLM was saying. It doesn’t imply that the OP thinks the LLM has internal beliefs matching that.

[awithrow]

Correct, it's really anthropomorphizing what gets generated. It's another form of the "you're absolutely right" framing. The interesting thing is sometimes it won't do that. It'll continue to insist that some point or another is still a valid interpretation. I'm some ways it all feels like a complicated way of reading tea leaves

[rsynnott]

Why not... do this with a person, instead? Other humans are available.

(Seriously, I don't understand this. Plenty of humans will be only too happy to argue with you.)

[kelseyfrog]

"the percentage of U.S. adults who report having no close friends has quadrupled to 12% since 1990"[1]

1. https://www.happiness.hks.harvard.edu/february-2025-issue/th...

[nathan_compton]

More technology is probably the solution to this!

[layla5alive]

Many other humans are .... Not very available - certainly many shut down when conversations reach a certain level of depth or require great focus or introspection..

[balamatom]

Depth? Introspection?

I'd say these days the norm is to not simply shut down, but to become irrevocably and insidiously hostile, the moment someone hints at the existence of such a thing as "ground truth", "subjective interpretation", "being right or wrong" - or any of the bits and bobs that might lead one to discover the proper scary notion, "consensus reality".

"What do you mean social reality is a constructed by the consensus of the participants? Reality is what has been drilled into my head under threat of starvation! How dare you exist!", et cetera. You've heard it translated into Business English countless times.

They are deathly afraid of becoming aware of their own conditioned state of teleological illiteracy - i.e. how they are trained to know what they are doing, but never why they are doing it. It's especially bad with the guys who cosplay US STEM gang.

One is not permitted a position of significance in this world without receiving this conditioning, and I figure it's precisely this global state of cognitive disavowal which props up the value of the US dollar - and all sorts of other standees you might've recently interacted with as if they're not 2D cutouts (metaphorical ones! metaphorical!).

PSA: Look up "locus of control" and "double bind". Between those two, you might be able to get a glimpse of what's going on - but have some sort of non-addictive sedative handy in case you do.

[VonTum]

You had me on the first three paragraphs, but the last two veer so far off course that I've no idea what you're trying to say. Mind clarifying?

[balamatom]

Yes

[gverrilla]

I think you will enjoy Guy Debord and Raoul Vaneigem.

[balamatom]

Just like a certain defense minister was shown to enjoy D&G; after which the latter were never heard of again. Where they go, eh?

+1 for Vaneigem, he has a nice cryptohistory of Nälkä; and you might also want to check out Villem Flusser.

[rsynnott]

> when conversations reach a certain level of depth or require great focus or introspection..

I mean... if the alternative is an LLM... you realise that the LLM isn't doing any focusing or introspection, right?

[matheusmoreira]

> Other humans are available.

Are they?

I have some personal projects that I enjoy working on whenever I have free time. One of them is a lisp interpreter. I just overhauled its memory allocator, now I'm working on the hash tables. Would you like to help me develop it?

[ip26]

No living breathing human deserves to be subjected to my level of overthinking, and vanishingly few share my fascination with my favorite topics.

[layer8]

In addition to availability, usually because you want to take advantage of the knowledge that is baked into the models, which for all its flaws still vastly exceeds the knowledge of any single human.

[emp17344]

For this use case, how do LLMs provide more value than a standard search engine? They may actually be destroying value here, as LLM-generated text pollutes search results.

[layer8]

They let you find things faster, and they combine and synthesize information from different sources. I’m more on the AI skeptic side and always prided myself on my Google foo, but nowadays chatbots can really save a lot of time with that.

Publishing LLM-generated text is a separate use case, I’m not a friend of that.

[awithrow]

oh i do as well. I think of the LLM as another tool in the toolbox, not a replacement for interactions. There is something different about having a rubber duck as a service though.

[mock-possum]

Arguing with a human costs social energy. Chatting with a robot does not.

[balamatom]

s/social/demonic/

[balamatom]

OK, I'll bite the artillery shell: I don't mean to dismiss you or what you are saying; in fact I strongly relate - wouldn't it be nice to be able to hash things out with people and mutually benefit from both the shared and the diverging perspectives implied in such interaction? Isn't that the most natural thing in the world?

Unfortunately these days this sounds halfway between a very privileged perspective and a pie in the sky.

When was the last time a person took responsibility for the bad outcome you got as a direct consequence of following their advice?

And, relatedly, where the hell do you even find humans who believe in discursive truth-seeking in 2026CE?

Because for the last 15 years or so I've only ever ran into (a) the kind of people who will keep arguing regardless if what they're saying is proven wrong; (b) and their complementaries, those who will never think about what you are saying, lest they commit to saying anything definite themselves, which may hypothetically be proven wrong.

Thing is, both types of people have plenty to lose; the magic wordball doesn't. (The previous sentence is my answer to the question you posited; and why I feel the present parenthesized disclaimer to be necessary, is a whole next can of worms...)

Signs of the existence of other kinds of people, perhaps such that have nothing to prove, are not unheard of.

But those people reside in some other layer of the social superstructure, where facts matter much less than adherence to "humane", "rational" not-even-dogmas (I'd rather liken it to complex conditioning).

But those folks (because reasons) are in a position of power over your well-being - and (because unfathomables) it's a definite faux pas to insist in their presence that there are such things as facts, which relate by the principles of verbal reasoning.

Best you could get out of them is the "you do you", "if you know you know", that sort of bubble-bobble - and don't you dare get even mildly miffed at such treatment of your natural desire to keep other humans in the loop.

AI is a symptom.

[rustystump]

I genuinely do not understand what u are saying. Because reasons, because unfathomables? Everyone in last 15 years has been an npc? I have had countless deep conversations with people and i am an uber introvert.

This reads like someone who is deep into their specific pov. You cannot hope to have a meaningful conversation if you yourself are not willing to concede a point.

To the op u are replying too, arguing with people can have real consequences if u say something stupid or carelessly. There is a another human there. With a machine, u are safe. At least u feel safe.

[hluska]

When you start hearing things like “you do you” or “if you know you know” it means that you went way too far. That’s a sign of discomfort.

If you make uncomfortable, you won’t get diverging perspectives. People will agree to anything to get out of a social situation that makes them uncomfortable.

If your goal is meaningful conversation, you may want to consider how you make people feel.

[balamatom]

Believe me (or don't), I always do. Even when this precludes a necessary conversation from happening. Even when the other party doesn't give a fuck about how they make others feel.

After all, if they're making me uncomfortable, surely there's something making them uncomfortable, which they're not being able to be forthright about, but with empathy I could figure it out from contextual cues, right?

>People will agree to anything to get out of a social situation that makes them uncomfortable.

That's fine as long as they have someone to take care of them.

In my experience, taking into account the opinions of such people has been the worst mistake of my life. I'm still working on the means to fix its consequences, as much as they are fixable at all.

"Doing whatever for the sake of avoiding mild discomfort" is cowardice, laziness, narcissism - I'm personally partial to the last one, but take your pick. In any case, I consider it a fundamentally dishonest attitude, and a priori have no wish to get along (i.e. become interdependent) with such people.

Other than that, I do agree with your overall sentiment and the underlying value system; I'm just not so sure any more that it is in fact correct.

[nuancebydefault]

> In my experience, taking into account the opinions of such people has been the worst mistake of my life. I'm still working on the means to fix its consequences, as much as they are fixable at all.

This sounds very cryptic. Can you give an example?

[balamatom]

Certainly. Can you guarantee my safety afterwards?

[balamatom]

Believe me (or don't), I always do. Even when this precludes a necessary conversation from happening. Even when the other party doesn't give a fuck about how they make others feel.

After all, if they're making me uncomfortable, surely there's something making them uncomfortable, which they're not being able to be forthright about, but with empathy I could figure it out from contextual cues, right?

>People will agree to anything to get out of a social situation that makes them uncomfortable.

That's fine as long as they have someone to take care of them.

In my experience, taking into account the opinions of such people has been the worst mistake of my life. I'm still working on the means to correct its consequences.

"Doing whatever for the sake of avoiding mild discomfort" is cowardice, laziness, narcissism - I'm personally partial to the last one, but take your pick. In any case, I see it as a way of being which is taught to people; and one which is fundamentally dishonest and irresponsible.

Other than that, I do agree with your overall sentiment and the underlying value system; I'm just not so sure any more that it is in fact correct.

[nuancebydefault]

Why is your wording so complicated? It is very hard for me to understand what you try to say, even though I am very interested.

[balamatom]

TL;DR: Probably because I'm having fun and you are expending effort. Hope you find what I say to be worth the effort.

To preface, I do not take offense to your remark, because you seem to be asking in good faith.

(If, however, being unable to immediately recognize pre-known patterns in my speech had automagically led you to the conclusion that I am somehow out of line, just for speaking how I speak ... well, then we woulda hadda problemo! But we don't, chill on.)

So, honest question deserves honest answer.

The short of it is: English sux.

Many many many people, much much much smarter than me (and much better compensated too!) have been working throughout modernity to make it literally impossible to express much of anything interesting in English.

(Well, not without either being a fictional character or sounding batshit insane, anyway! But that joke's entirely on "the Them": I am not only entirely fictional, but have an equal amount of experience being batshit insane in my native language and in the present lingua franca. So, consider all I say cognitohazardous and watch out for colors you ain't seen before, dawg!)

Linguistic hegemony is the thing that LLMs are the steroids for - surfuckingprise! - and that's why your commanders love 'em.

As opposed to programming languages, which your superiors loathe and your peers viscerally refuse to acknowledge, because those are the exact opposite thing: descending from mathemathical notation, and being evaluated by a machine, they have the useful property of being incapable of expressing lies and nonsense.

Direct computing confers what you could call bullshit-resistance. That property is a treasure underappreciated by virtue of its unfamiliarity, and one which we are in the process of being robbed of.

I also want to admit that linguistic hegemony isn't all downside: English is great for technical and instrumental knowledge - especially with elided bells and whistles (adverbs, copula, etc.)

But then life ain't all business, izzet?

Imagine you have a partner who wants to have a conversation about feelings and interpersonal relations; and not even in a scary way, right? So you sit and talk about stuff, and your partner does this thing where they keep switching from your shared native tongue to English mid-sentence, in order to be able to talk about such things better, because your native tongue does not have - no, not only the established words and notions! - it doesn't have the basic grammatical constructs for expressing simple things unambiguously, so if you were to attempt the same conversation in nativelang you'd end up battling it out with proverbs and anodyne canards ripped from propaganda repertoire of the prior regime.

Fun, no?

As an exercise, try imagining what notions are absent from modern English. And don't forget to remain vigilant. Love from our table to your table!

[nuancebydefault]

You lost me well before the anodyne canards...

When talking about feelings, we now and then throw in an English word because some things are expressed in much less words when using English. In a few cases even a German word. Überhaupt is for example a word for which i do not know an alternative in any language.

I think you want to say that human language is too ambiguous for clear communication between human and machine. The machine might mis interpret what you write. Classic computer languages leave no room for interpretation.

For the rest I think you might be a little lost. That is okay, so many of us are. I wish you all the best.

[balamatom]

>I think you want to say that human language is too ambiguous for clear communication between human and machine. [...]

If that is what I wanted to say, I figure I would not have had much difficulty with saying exactly it - and not something else.

Except I fail to see the purpose of making that statement.

Maybe to have some people say "it is true! I agree with what the balamatom is saying"?

Again - to what end? How would that agreement be of use to me?

Why say something which both speaker and listener have already heard a thousand times? To get a cracker and be called pretty?

And have I lost the author, or have I lost the reader, or we all so lost that it doesn't matter how lost each is? Maybe one day we will all become so lost that it will once again begin to matter where exactly we are! Counting on it.

[rsynnott]

> When was the last time a person took responsibility for the bad outcome you got as a direct consequence of following their advice?

... I mean, the LLM certainly isn't going to do that.

[balamatom]

Yes, exactly - while humans are "only" highly unlikely to do so.

Wasn't that the premise?

There's X% probability to get unaccountable bad advice from a human; Y% probability of same from LLM.

But the stakes are different not because X!=Y. In fact, those probabilities might as well be equal. It's not like one can do much more than vaguely intuit them.

The stakes are different because the LLM is not considered a person, and therefore, one risks less by asking it a question.

(Ostensibly, anyway.)

>Other humans are available.

Sure. Available to miss my point entirely. And then to be upset by my attempts to point it out more precisely. And then downhill from there.

[magicalhippo]

Gemini seems to be fairly good at keeping the custom instructions in mind. In mine I've told it to not assume my ideas are good and provide critique where appropriate. And I find it does that fairly well.

[steve_adams_86]

Same. This works fine for Claude in my experience. My user prompt is fairly large and encourages certain behaviours I want to see, which involves being critical and considering the strengths and weaknesses of ideas before drawing conclusions. As someone else mentioned, there does seem to be a phenomenon where saying DO NOT DO X causes a sort of attention bias on X which can lead to X occurring despite the clear instructions. I've never empirically tested that, I've just noticed better results over the years when telling it what paths to stick to rather than specific things not do to.

[koverstreet]

That happens with humans too :) It's why positive feedback that draws attention to the behavior you want to encourage often works better. "Attention" is lower level and more fundamental than reasoning by syllogism.

[magicalhippo]

> there does seem to be a phenomenon where saying DO NOT DO X causes a sort of attention bias on X which can lead to X occurring despite the clear instructions

It's a thing with people too[1], ie do not think about a white bear.

[1]: https://en.wikipedia.org/wiki/Ironic_process_theory

[steve_adams_86]

Don't think about what?

[iugtmkbdfil834]

I will admit that I was very pleasantly surprised by gemini lately. I was away from my PC and tried it on a whim for a semi-random consumer question that led into smaller rabbit hole. It seemed helpful enough and focused on what I tried to get while still pushing back when my 'solutions' seemed out of whack.

[lelanthran]

> Gemini seems to be fairly good at keeping the custom instructions in mind.

Unless those instructions are "stop providing links to you for every question ".

[magicalhippo]

I think that's part of the RAG pipeline, at least to me it looks to be separate from the model output. Models are notorious for getting URLs wrong so makes sense to me to handle it separately.

[Loughla]

That's because you need actual logic and thought to be able to decide when to be critical and when to agree.

Chatbots can't do that. They can only predict what comes next statistically. So, I guess you're asking if the average Internet comment agrees with you or not.

I'm not sure there's much value there. Chatbots are good at tasks (make this pdf an accessible word document or sort the data by x), not decision making.

[kvirani]

I'm not convinced that "actual logic and thought" aren't just about inferring what comes next statistically based on experience.

[dinkumthinkum]

Is this just Internet smart contrarianism or a real thing? Are logic gates in a digital circuit just behaving statistically according to their experience?

[hluska]

You know, you might really enjoy consumer behaviour. When you get into the depths of it, you’ll end up running straight into that idea like you’re doing a 100 metre dash in a 90 metre gym. It’s quite interesting how arguably the best funded group under the psychology umbrella runs directly into this. One of my favourite examples is how heuristics will lead otherwise reasonable people to make decisions that are not in their interest.

[righthand]

Communicating is usually about inferring. I dont think token to token. And I don’t think “well statistically I could say ‘and’ next but I will say ‘also’ instead to give my speech some flash”. If I decided on swapping a word I would have made my decision long ago, not in the moment. Thought and logic are not me pouring through my brain finding a statistical path to any answer. Often I stop and say “I dont know”.

[Swizec]

> I'm not convinced that "actual logic and thought" aren't just about inferring what comes next statistically based on experience.

Often they are the exact opposite. Entire fields of math and science talk about this. Causation vs correlation, confirmation bias, base rate fallacy, bayesian reasoning, sharp shooter fallacy, etc.

All of those were developed because “inferring from experience” leads you to the wrong conclusion.

[theptip]

Bayesian reasoning is just another algorithm for predicting from experience (aka your prior).

I took the GP to be making a general point about the power of “next x prediction” rather than the algorithm a human would run when you say they are “inferring from experience”. (I may be assuming my own beliefs of course.)

Eg even LeCun’s rejection of LLMs to build world models is still running a predictor, just in latent space (so predicting next world-state, instead of next-token).

And of course, under the Predictive Processing model there is a comprehensive explanation of human cognition as hierarchical predictors. So it’s a plausible general model.

[Swizec]

> under the Predictive Processing model there is a comprehensive explanation of human cognition as hierarchical predictors

It’s plausible!

But keep in mind humans have been explaining ourselves in terms of the current most advanced technology for centuries. We used to be kinda like clockwork, then a bit like a steam engine, then a lot like computers, and now we’re just like AI.

That’s why you blow a gasket or fuse, release some steam, reboot your life, do brain dump, feel like a cog in the machine, get your wires crossed, etc

[plagiarist]

Then the machines still need a more sophisticated "experience" compared to what they have currently.

[theptip]

Exactly. Lots can be explained just with more abstract predictors, plus some mechanisms for stochastic rollout and memory.

[righthand]

I said this pretty much and got major downvotes…

[dTal]

Because it's an outmoded cliche that never held much philosophical weight to begin with and doesn't advance the discussion usefully. "It's a stochastic parrot" is not a useful predictor of actual LLM capabilities and never was. Last year someone posted on HN a log of GPT-5 reverse engineering some tricky assembly code, a challenge set by another commentator as an example of "something LLMs could never do". But here we are a year later still wading through people who cannot accept that LLMs can, in a meaningful sense, "compute".

[righthand]

It’s entirely useful discussion because as soon as you forget that it’s not really having a conversation with you, it’s a deep dive into delusion that you’re talking to a smart robot and ignoring the fact that these smart robots were trained on a pile of mostly garbage. When I have a conversation with another human, I’m not expecting them to brute force an answer to the topic. As soon as you forget that Llms are just brute forcing token by token then people start living in fantasy land. The whole “it’s not a stochastic parrot” is just “you’re holding it wrong”.

[layla5alive]

Its not that LLMs are stochastic parrots and humans are not. Its that many humans often sail through conversations stochastic parroting because they're mentally tired and "phoning it in" - so there are times when talking to the LLM, which has a higher level of knowledge, feels more fruitful on a topic than talking to a human who doesn't have the bandwidth to give you their full attention, and also lack the depth and breadth of knowledge. I can go deep on many topics with LLMs that most humans can't or won't keep up on. In the end, I'm really only talking to myself most of the time in either case, but the LLM is a more capable echo, and it doesn't tire of talking about any topic - it can dive deep into complex details, and catching its hallucinations is an exercise in itself.

[dinkumthinkum]

No. It's quite a useful thing to understand So, what, you have us believe it is a sentient, thinking, kind of digital organism and you would have us not believe that it is exactly what it is? Being wrong and being unimaginative about what can be achieved with such a "parrot" is not the same as being wrong about it be a word predictor. If you don't think, you can probably ask an LLM and it will even "admit" this fact. I do agree that it has become considered to be outmoded to question anything about the current AI Orthodox.

[plagiarist]

People are upset hearing that LLMs aren't sentient for some reason. Expect to be downvoted, it is okay.

[gjm11]

First off, "not adequately described as a mere token-predictor" and "not sentient" are entirely separate things.

I can't speak for anyone else, but what I feel when I read yet another glib "it's just a stochastic parrot, of course it isn't doing anything that deserves to be called reasoning" take is much more like bored than it is like upset.

Today's LLMs are in some sense "just predicting tokens" in some sense. Likewise, human brains are in some sense "just shuttling neurotransmitters and electrical impulses around" in some sense. Neither of those tells you what the thing can actually do. To figure that out, you have to look at what it can do.

Today's best LLMs can do about as well as the best humans on problems from the International Mathematical Olympiad and occasionally solve easyish actual mathematical research problems. They write code about as well as a junior software developer (better in some ways, worse in others) but much faster. They write prose about as well as an average educated person (but with some annoying quirks that are annoying mostly because they are the same quirks over and over again).

If it pleases you to call those things "thinking" then you can. If it pleases you to call them "stochastic parroting" then you can. They are the same things either way. They are not, on the face of it, very much like "just repeating things the machine has already seen", or at least not more like that than a lot of things intelligent human beings do that we don't usually describe that way.

If you want to know whether an LLM can do some particular thing -- do your job well enough for your boss to fire you, write advertising copy that will successfully sell products, exterminate the human race, whatever -- then it's not enough to say "it's just remixing what it's seen on the internet, therefore it can't do X" unless you also have good reason to believe that that thing can't be done by just "remixing what's on the internet" (in whatever sense of "remixing" the LLM is doing that). And it's turning out that lots of things can be done that way that you absolutely wouldn't have predicted five years ago could be done that way.

It seems to me that this should make us very cautious about saying "they can't do X because all they can do is regurgitate a combination of things they've seen in training".

(My own view, not that there's any reason why anyone should care what I-in-particular think, is a combination of "what they're doing is less parroting than you might have thought" and "you can do more by parroting than you might have thought".)

So, anyway, this particular instance of the stochastic-parrot argument started when someone said: of course the AIs are yes-men, because figuring out when to agree and when not to requires actual logic and thought and the LLMs don't have either of those things.

Is it really clear that deciding whether or not to agree when someone says "I think maybe I should break up with my girlfriend" or "I've got this amazing new theory of physics that the establishment is stupidly dismissing" requires more logic and thought than, say, gold-medal performance on IMO problems? It certainly isn't clear to me. Having done a couple of International Mathematical Olympiads myself in my tragically unmisspent youth, I can assure you that solving their problems requires quite a bit of logic and thought, at least for humans. It may well be harder to give a good answer to "should I leave my job?", but it's not exactly "logic and thought" that it needs more of.

Someone reported that Claude is much less yes-man-ish than Gemini and ChatGPT. I don't know whether that's true (though it wouldn't surprise me) but: suppose it is; do you want that to oblige you to say that yes, actually, Claude really thinks logically, unlike Gemini and ChatGPT? I don't think you do. And if not, you want to avoid saying "duh, of course, you can't avoid being a yes-man without actually thinking and reasoning, and we all know that LLMs can't do those things".

[rustystump]

I wont touch how profoundly i disagree with everything you said on reasoning (u clearly already have it figured out) but a fun test i have done with most of the big models is to give it some text input, maybe a short story, and have it rate it. That is, the prompt is, rate this from 1-10.

For Gemini and gpt, it almost always will give very similar scores for everything. As long as grammar isnt off u cannot get below a 7.

X ai on the other hand will rarely give anything above a 7.

Now when u prompt with, rate 1-10 with 5 being average, all the sudden the scores of openai and gemini drop and x ai remains roughly the same.

All of them will eventually give you a 10 if u keep making tiny edits “fixing” whatever they complain about.

Humans do not do this. Or more specifically, my experience with humans.

[gjm11]

Clearly a bunch of other people also disagree profoundly with everything I said, since my comment is currently sitting at 0 having at one point been higher.

I vigorously encourage anyone who thinks something I wrote is bad to downvote it as they see fit, but it would be nice if some of those people would tell me what about my comment they found so objectionable. (It all seems pretty well reasoned to me -- but it would, wouldn't it?)

[EDITED to fix an inconsequential typo]

[secret_agent]

Use positive requests for behavior. For some reason, counter prompts "Don't do X" seems to put more attention on X than the "Don't do." It's something like target fixation, "Oh shit I don't want to hit that pothole..." bang

[ambicapter]

This is a well known problem in these kind of systems. I’m not 100% on what the issue is mechanically but it’s something like they can only represent the existence of things and not non-existence so you end up with a sort of “don’t think of the pink elephant” type of problem.

[SpicyLemonZest]

Isn't it just that, in the underlying text distribution, both "X" and "don't do X" are positively correlated with the subsequent presence of X? I've never seen that analysis run directly but it would surprise me if it weren't true.

[cyanydeez]

So, there's things you're fighting against when trying to constrain the behavior of the llm.

First, those beginning instructions are being quickly ignored as the longer context changes the probabilities. After every round, it get pushed into whatever context you drive towards. The fix is chopping out that context and providing it before each new round. something like `<rules><question><answer>` -> `<question><answer><rules><question>`.

This would always preface your question with your prefered rules and remove those rules from the end of the context.

The reason why this isn't done is because it poisons the KV cache, and doing that causes the cloud companies to spin up more inference.

[colechristensen]

>"Help me refine my ideas, challenge, push back, and don't just be agreeable."

This is where you're doing it wrong.

If your LLM has a problem being more agreeable than you want, prompt it in a way that makes being agreeable contrary to your real intentions.

"there are bugs and logic problems in this code" "find the strongest refutation of this argument" "I don't like this plan and need to develop a solid argument against it"

Asking for top ten lists is a good method, it will rarely not come up with anything but you can go back and forth and refine until it's 10 ten reasons why your plan is bad are all insubstantial nonsense then you've made progress

[raincole]

My rule of thumb:

1. Only one shot or two shot. Never try to have a prolonged conversation with an LLM.

2. Give specific numbers. Like "give me two alternative libraries" or "tell me three possible ways this might fail."

[RugnirViking]

check out this article that was posted here a while back https://www.randalolson.com/2026/02/07/the-are-you-sure-prob...

The article's main idea is that for an AI, sycophancy or adversarial (contrarian) are the two available modes only. It's because they don't have enough context to make defensible decisions. You need to include a bunch of fuzzy stuff around the situation, far more than it strictly "needs" to help it stick to its guns and actually make decisions confidently

I think this is interesting as an idea. I do find that when I give really detailed context about my team, other teams, ours and their okrs, goals, things I know people like or are passionate about, it gives better answers and is more confident. but its also often wrong, or overindexes on these things I have written. In practise, its very difficult to get enough of this on paper without a: holding a frankly worrying level of sensitive information (is it a good idea to write down what I really think of various people's weaknesses and strengths?) and b: spending hours each day merely establishing ongoing context of what I heard at lunch or who's off sick today or whatever, plus I know that research shows longer context can degrade performance, so in theory you want to somehow cut it down to only that which truly matters for the task at hand and and and... goodness gracious its all very time consuming and im not sure its worth the squeeze

[cruffle_duffle]

> goodness gracious its all very time consuming and im not sure its worth the squeeze

And when you step back you start to wonder if all you are doing is trying to get the model to echo what you already know in your gut back to you.

[awithrow]

oh that's great. thanks for the link!

[oldfrenchfries]

This is great, thanks for sharing!

[ajkjk]

'admit' isn't really the right word for that... the fact that it was placating you wasn't true until you prompted it to say so. Unlike a person who has an 'internal emotional state' independent of what they say that you can probe by asking questions.

[awithrow]

'admit' is anthropomorphizing the behavior, sure. The point is that sometimes the model's response will tighten, flag things that were overly supportive or what not. Sometimes it wont, it'll state that previous positions are still supported and continue to press it. Its not like either response is 'correct' but it can alter the rest of the responses in ways that are useful.

[GuB-42]

I find the best way is to give the LLM as little information as possible about where you want to go. For example don't say "I think pineapple pizzas are the best, am I right?", say "What is the general consensus on pineapple pizzas?".

[sinuhe69]

The longer you converse with the LLM, the more frequent the same keywords will occur and Thu’s they will only strengthen the probabilistic choices of the LLM. The LLM do not exactly follow the instructions like a program would do, nor does it truly understand its role or what its “said”. It looks fine on first tries and short texts, but will drift more the longer the conversation goes. The only way the LLM can maintain the appearance of a normal conversation is to repeatedly feed it with the users requests. We can see it with the shell construct the AI company built around their chat interface.

[jimmyjazz14]

Yeah, I have never had good results with refining ideas with models or really any interactions with models outside of rote task such as coding or analyzing document structures, I don't know why I was ever surprised by this as its obvious that LLMs just aren't capable of original thinking. I think part of the problem is that these things were marketed originally as chatbots when that is honestly their weakest use-case. I think even when I was expressly try to not anthropomorphize LLMs I still sorta did in early days, but the less I do so the more utility I get from them.

[arjie]

I haven't found that work at all. But the way I usually frame it for this is that I present it as someone else's work that I disagree with (sometimes "this is a junior engineer's work and he often does dumb shit" as an introduction to an essay). In this way, it tries to be agreeable with me and can rationalize various opposition. And then I can select from that.

[Forgeties79]

I usually put “do not praise me, do not use emojis, I just want straight answers” something along those lines and it’s been surprisingly effective. Though it helps I can’t run particularly heavy duty models/don't carry on the “conversation” for super long durations.

[margalabargala]

Considering 4.6 came with a ton of changes around tooling and prompting this isn't terribly surprising.

[dkersten]

I find Kimi white good if you ask it for critical feedback.

It’s BRUTAL but offers solutions.

[awithrow]

what is Kimi white?

[dkersten]

I was typing quickly on my phone. I meant “quite”, “I find Kimi quite good”

[ohyoutravel]

Not soft, not mild, but BRUTAL! This broke my brain!

[anandram27]

Could be an aspect of eval awareness mb

[dinkumthinkum]

You're not wrong and you're not crazy. In fact, you are absolutely right! It is not just These things are not just casual enablers. They are full-on palace sycophants following the naked emperor showering him with praise for his sartorial elegance. /s

[righthand]

That’s because the model isn’t actually thinking, pushing back, and challenging your ideas. It’s just statistically agreeing with you until it reaches too wide of a context. You’re living in the delusion that it’s “working” or having a “conversation” with you.

[alehlopeh]

How is conceptualizing what the model is doing as having a conversation any different from any other abstraction? “No, the browser isn’t downloading a file. The electrons in the silicon are actually…”

[colechristensen]

There are people with a philosophical objection to using everyday words to describe LLM interactions for various reasons, but commonly because they're worried stupid people will confuse the LLM for a person. Which, I suppose stupid people will do that, but I'm not inventing a parallel language or putting a * next to each thing which means "this, but with an LLM instead of a person"

[cruffle_duffle]

That is an interesting way of looking at that, thanks for the perspective!

Like, the words fit… why create a second parallel language for describing LLM behavior.

Somebody else said it… the whole “it’s a stochastic parrot” thing is sooooo cliche and boring at this point. It’s like, duh… what is your point?

[righthand]

One is an abstraction, another is a fantasy.