site banner

Culture War Roundup for the week of August 11, 2025

This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.

Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.

We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:

  • Shaming.

  • Attempting to 'build consensus' or enforce ideological conformity.

  • Making sweeping generalizations to vilify a group you dislike.

  • Recruiting for a cause.

  • Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.

In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:

  • Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.

  • Be as precise and charitable as you can. Don't paraphrase unflatteringly.

  • Don't imply that someone said something they did not say, even if you think it follows from what they said.

  • Write like everyone is reading and you want them to be included in the discussion.

On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.

3
Jump in the discussion.

No email address required.

Training language models to be warm and empathetic makes them less reliable and more sycophantic:

Artificial intelligence (AI) developers are increasingly building language models with warm and empathetic personas that millions of people now use for advice, therapy, and companionship. Here, we show how this creates a significant trade-off: optimizing language models for warmth undermines their reliability, especially when users express vulnerability. We conducted controlled experiments on five language models of varying sizes and architectures, training them to produce warmer, more empathetic responses, then evaluating them on safety-critical tasks. Warm models showed substantially higher error rates (+10 to +30 percentage points) than their original counterparts, promoting conspiracy theories, providing incorrect factual information, and offering problematic medical advice. They were also significantly more likely to validate incorrect user beliefs, particularly when user messages expressed sadness. Importantly, these effects were consistent across different model architectures, and occurred despite preserved performance on standard benchmarks, revealing systematic risks that current evaluation practices may fail to detect. As human-like AI systems are deployed at an unprecedented scale, our findings indicate a need to rethink how we develop and oversee these systems that are reshaping human relationships and social interaction.

Assuming that the results reported in the paper are accurate and that they do generalize across model architectures with some regularity, it seems to me that there are two stances you can take regarding this phenomenon; you can either view it as an "easy problem" or a "hard problem":

  • The "easy problem" view: This is essentially just an artifact of the specific fine-tuning method that the authors used. It should not be an insurmountable task to come up with a training method that tells the LLM to maximize warmth and empathy, but without sacrificing honesty and rigor. Just tell the LLM to optimize for both and we'll be fine.

  • The "hard problem" view: This phenomenon is perhaps indicative of a more fundamental tradeoff in the design space of possible minds. Perhaps there is something intrinsic to the fact that, as a mind devotes more attention to "humane concerns" and "social reasoning", there tends to be a concomitant sacrifice of attention to matters of effectiveness and pure rigor. This is not to say that there are no minds that successfully optimize for both; only that they are noticeably more uncommon, relative to the total space of all possibilities. If this view is correct, it could be troublesome for alignment research. Beyond mere orthogonality, raw intellect and effectiveness (and most AI boosters want a hypothetical ASI to be highly effective at realizing its concrete visions in the external world) might actually be negatively correlated with empathy.

One HN comment on the paper read as follows:

A few months ago I asked GPT for a prompt to make it more truthful and logical. The prompt it came up with included the clause "never use friendly or encouraging language"

which is quite fascinating!

EDIT: Funny how many topics this fractured off into, seems notable even by TheMotte standards...

Whatever Anthropic does with Claude seems to work. It's the most flavorful model without really trying too hard to be bubbly and quirky like GPT-4o. Of course, it has its own sycophancy issues, but nowhere near as bad as 4o. (The least sycophantic model I know is Kimi K2, which is incredibly cynical, which makes it interesting)

I am more inclined to go with the "easy problem" view, or perhaps a halfway position. Sycophancy isn't an insurmountable problem. If you're not careful, then trying to knock out obvious sycophancy will make the model more prone to looking for ways to subtly achieve the goal of tricking/convincing the user into giving positive feedback.

To a degree, we really must ask ourselves what "warm and empathetic" really means:

  • If a five year old child asks for feedback on an essay, it is arguably almost always true that their writing sucks. That might be true, but it is a tad-bit unhelpful. The most socially adept/instrumentally useful answer (without outright lying) is to praise them for the effort, offer improvements, and tell them to keep at it. Of course, if you're in literary masters course, and the exact same standard of writing is presented before you, some more colorful verbiage might be appropriate.

  • A lot of social interaction is lubricated by white lies, and a lot of what is deemed "politeness" isn't maximally truth-seeking.

Perhaps maximal truth-seeking conflicts with warmth and empathy. It's possible the tails come apart. But I don't think they're outright opposed to each other, and you can probably find a Pareto frontier that makes most people happy.

Current Claude is all right, but 3.5 (or was it 3.6? I'm forgetting already..) was best Claude. Its defining attribute was that it was relentlessly curious. That felt empathetic, yet truth-seeking without being sycophantic.

But apparently it sucked at code, so it was taken out back and shot :(

AI companies all fail at naming things. There was:

  • Claude 3.5 (June 2024)
  • Claude 3.5 (October 2024)
  • Claude 3.7 (Feb 2025)

You're probably thinking of the one between the original 3.5 and 3.7.