A Superficial Aesthetic of Moderation

[osteophage]

Overvaluing the appearance of moderation can be a mistake, in that it can give undue credence to assertions that go unsubstantiated. This pitfall can be illustrated by analyzing arguments that position themselves as moderate by making assertions with little to no supporting evidence. While offhand superficial assertions are not themselves a sin, they deserve to be recognized for what they are, rather than held in high esteem based on nothing more than an aesthetic—a superficial aesthetic of moderation.

To demonstrate the concept, this post presents many examples. The first section identifies some otherwise generally solid arguments that position themselves as moderate via the insertion of superficial assertions. For contrast, the second section highlights an argument that actually substantiates the kind of claim left unsubstantiated in the previous section. The mere appearance of a moderate stance is no substitute for substance, and when evaluating the quality of an argument, it’s important to look for that substance rather than relying on aesthetics as a shortcut. An example in the final section serves to illustrate why.

Note: this post draws examples from the LLM debate, but the goal here is to advance a more general principle. For more direct discussion of LLMs themselves, see this post or these links.

Crossposted to Pillowfort and my personal site. 

Examples of Superficial Assertions

Here are six different posts that demonstrate a certain recurring asymmetry. Each one discusses LLMs in terms of pros and cons. However, one side is presented with substantial support, while the other is presented in a way that is left comparatively superficial. That superficiality can involve making claims without any evidence, neglecting to name any specific examples, or using examples that come across as very thin and misaligned with other parts of the same post.

For the sake of starting off on a high note, we can begin with When All You Have Is a Robots.txt Hammer by Nick Heer, which is a response to a post by Bluesky guy Mike Masnick. Mike’s post is, in a word, bad. In response, Nick Heer very delicately presents some counterpoints. For our purposes here though, consider this offhand remark toward the very end:

Regardless of the benefits they have created — and I do believe there are benefits to these technologies — they have behaved unethically.

Benefits to these technologies have not been indicated or described in this post, a choice which you might chalk up to a matter of intended audience. Even so, the remark represents an instance of a larger pattern, and this pattern recurs even when there is no such basis to assume a shared understanding.

LLMs are bullshitters, but that doesn’t mean they’re not useful by Matt Ranger, which is essentially a piece of marketing for Kagi, sports a title that would, in theory, be suited to a post all about explaining how LLMs can be useful. Instead it features the statement “I don’t think I need to list the large number of tasks where LLMs can save humans time, if used well.” A few lines later, Matt adds, “By all means, use LLMs where they are useful tools: tasks where you can verify the output, where speed matters more than perfection, where the stakes of being wrong are low.” Which tasks fit the bill? Well, the specific tasks that get alluded to in this section are “help me find a certain page in a document” and “sanity check this post.”

Colin’s Generative AI Threatens Diversity and Hyperlinks, which is effectively a piece of marketing for Mojeek, plays out much the same way. This post begins with the line “There is no doubt that generative AI and chatbots can be useful.” Since it presumes no doubt, it provides no evidence. The bulk of the piece actually focuses on the downsides. Later on, another paragraph begins, “We are not suggesting that generative AI and chatbots are not useful, when used appropriately.” What it means for them to be “used appropriately” goes unexplained.

Even posts with more acerbic titles have ended up participating in this pattern, such as The I in LLM stands for Intelligence by Daniel Stenberg. In this post Daniel states that “AI can do a lot of good things,” but to get the full effect, take a look at this statement in context:

I realize AI can do a lot of good things. As any general purpose tool it can also be used for the wrong things. I am also sure AIs can be trained and ultimately get used even for finding and reporting security problems in productive ways, but so far we have yet to find good examples of this.

The bulk of this post, again, is about how LLMs can be detrimental. Amid all that, the “good things” remark, in its brevity, goes unsubstantiated—written in the tone of someone preempting a counterpoint, presented and then immediately pivoted away from.

For comparatively more substance on that front, you can look to Evaluating AI by Ben Werdmuller. Ben’s post, which is packed with criticism and evidence for that criticism, makes sure not to position itself as entirely opposed; Ben says that LLMs “can be used to augment [people’s] work,” says they are “potentially useful tools to speed up someone’s work,” and asserts that “there are uses for AI, and it should be included as part of a modern toolbox.” So what are those uses? A section under the heading of “So how can I use AI?” finally presents some answers:

The answer to the question “how can I use AI” is the same as the answer to how you should approach using any technology: carefully, and with a strong handle on your needs and values.

It’s the classic dad joke in action. How can I use these things? Very carefully.

To be fair, though, Ben’s post does provide some specific examples. For instance, later in this same section, he suggests that an LLM can “give you an overview of what to prepare for a meeting” or “summarize things you need to understand.” This is where it would make sense to provide evidence that LLMs can summarize things reliably. Instead, Ben links to a news article about them summarizing unreliably.

In the examples above, this recurring style of positive-but-surface-level statement seems geared toward softening what might otherwise appear to be a more critical stance, which would fit the advice presented in How the Anti-AI Movement Hurts Itself (and What It Could Do Instead) by Jens Oliver Meiert. Jens grants that critics have made some important points, and he asks that people try to make those points “without turning into jerks.” For instance, he objects to critics neglecting to “acknowledge even one positive aspect of AI.” The nature of these “positive aspects” goes unspecified. Instead of illuminating anything in particular on that front, Jens recommends that critics should “find a spot on the spectrum” and try to avoid looking “extremist.”

Put a pin in that. We’ll come back to it.

Example of a More Substantive Argument

For an example of what it looks like to pair more moderate positioning with more substantial support, look to AI isn’t useless, but is it worth it? by Molly White. This post positions itself as moderate in that it both criticizes LLMs and also distances itself from the more absolutist claim that LLMs are “useless.” In disputing the latter, the post illustrates some specific use cases where an LLM succeeded at a useful task, such as identifying a related word or generating a Python script for converting an OPML file. These specific examples are presented as direct counterpoints to the “useless” characterization, in order to challenge critics to be more precise in their choice of words.

Molly’s post is worth contrasting with others because her post illustrates what the others are lacking. Rather than treating an assertion like “LLMs aren’t useless” as automatically accepted and safe to assume, she does the work of presenting specific and detailed supporting evidence. In other words, this is an argument with substance.

Learning how to spot the substance is an important part of evaluating any given argument, whether on this subject or any other, and here’s why:

The Treachery of Vibes

Overvaluing an aesthetic of moderation can be treacherous. Part of its appeal may lie in the expectation that a moderate stance arises from a cautious, level-headed, well-reasoned approach, but mistaking the former for the latter can be a trap. Where the two are conflated, superficial assertions can skate by undetected, which means they can mislead people into buying ideas that don’t really hold up to scrutiny.

For example, last year, Jeremy Keith reflected on initially having a high estimation of a seemingly “fair and balanced” post, before later realizing that this appearance was illusory. The post in question was Is it okay? by Robin Sloan, which was met with criticism by Baldur Bjarnason. In the ensuing discussion, Jeremy spotted a disparity in how Robin and Baldur were each approaching the subject:

Baldur kept bringing the receipts. That’s when it struck me that Robin’s stance is largely based on vibes, whereas Baldur’s viewpoint is informed by facts on the ground.

I really appreciate that Jeremy reassessed his initial outlook this way.

In light of that, his experience should illuminate the hazards embedded in Jens’ advice to pursue an aesthetic of moderation:

Find a spot on the spectrum. There simply aren’t only and exclusively problems with AI. Insisting on that may look principled to AI critics, but it appears extremist and fanatic to anyone who has ever solved a task with AI.

This advice might superficially appear like the same advice presented by Molly White, but it does not attune itself to the same specificity. Molly’s post targets a specific type of claim with specific counterpoints in order to discourage people from going out on a limb with assertions that are easily contradicted. In contrast, the directive in the quote above is oriented more abstractly toward appearances: take care not to appear “extremist and fanatic.” Instead, aim to position yourself as a moderate. Prioritize an aesthetic of moderation.

While an aesthetic of moderation may well be preferred by some readers, the directive to prioritize it does not take into account how this aesthetic preference could itself be dangerous or worthy of criticism. What do we lose out on by committing ourselves to an aesthetic of moderation? Is it necessarily worth the tradeoff? This concern extends to more than just the LLM debate. Consider how this dynamic has already been scrutinized time and time again:

As ever with these sorts of articles, there’s a profound unwillingness to actually engage with an argument unless it’s in abstract terms, and an equally profound dislike of “ethical” discussion that asserts you have to choose a side.

—This Day’s Portion, On Not Choosing Nice Versions of AI (2025)

The thing is, the world is on fire now, and I can’t help but worry how much of this is a harbinger of the next four years—or, God help us, decade or longer. There’s going to be pressure to not take strong stands. […] There’s going to be, in other words, pressure to be genteel. That means not calling people out when they deserve to be called out. It means looking the other way. Maybe there are times in history where that’s fine, or at least not harmful, but we’re not in such a time.

—Watts Martin, The Pressure to Stay Genteel (2025)

Over time, it has become clear that anti-negativity is a worldview of its own, a particular mode of thinking and argument, no matter how evasively or vapidly it chooses to express itself. […]
To actually say a plain and direct word like “corrupt” is more outlandish, in smarm’s outlook, than even swearing.

—Tom Scocca, On Smarm (2013)

It’s nearly impossible to be “polite” while talking about race. The deck is stacked. It’s not a conversation “polite” people have. There is a conception of what “polite” is, and it doesn’t involve confronting people on their racism. Merely bring up racism is sufficient to make the conversation “impolite.” Now I put those in scare quotes because I acknowledge that that version of “polite” is different from attempting to show respect to fellow human beings, but it’s there, and quite often appeals to politeness are in fact demands to shut up.

—Brown Betty, commenting on The “Tone” Argument (2009)

An aesthetic of polite, genteel, nonthreatening moderation is nothing more than an aesthetic. It can at times be paired with very superficial assertions, and while superficial assertions are not inherently a bad thing, they should not be treated as categorically exempt from scrutiny just because of their tone. There is an important distinction to be made between due caution and premature concession. So, in a nutshell:

Beware the treachery of vibes.

Comments

[zavodilaterrarium]

The call to merely appear moderate is so frustrating, both to witness and experience. Had an acquaintance who I decently liked (it is of note that they had no connection to anyone else I knew so it was low stakes), enough that I inadvertantly didn't address the biggest problems I had with their beliefs. Cut them off once they made it a "agree to disagree" issue since it just wasn't worth it, and soon after realised that I regretted not having enough of a backbone from the very start just because I didn't want to come off 'aggressive'. I don't even hold my tongue particularly with my parents, whom I heavily rely on but do quite disagree with at times, I absolutely should not have bothered being agreeable with an acquaintance just for the sake of it!

[osteophage]

It can be tough to navigate those confrontations no matter what, really. I'm sorry you had that experience.