site banner
PaperclipPerfector  1mo ago (text post)   3285 thread views

Small-Scale Question Sunday for August 16, 2026

Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?

This is your opportunity to ask questions. No question too simple or too silly.

Culture war topics are accepted, and proposals for a better intro post are appreciated.

3
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

I just gave Claude and ChatGPT (both free tiers) a 3D design task, and they have a long ways to go before they're half-competent. The general shape for both was:

  1. I have a countertop dishwasher, and I need a more secure way to hold its drainage hose so it dumps in the sink. I can 3D print something if you design it.
  2. They both quickly chose to make a clamp from the front edge of the countertop to the inside of the sink, to hold the hose in place.
  3. They spent an hour giving me terrible STLs. There were hovering pieces, assembled pieces that overlapped with each other or the counter/sink, connections that simply butted up to each other instead of attaching in any way, orientation issues, and blocking the hose path. We didn't even get to printability.

For reference, it took me about five minutes to go from noticing the problem to starting the print (the first design worked, no additional iterations). Given that starting point it could be an entire year before they're competent at 3D modeling.

You really can't conclude anything from trying the free tiers. The difference between Sonnet and Fable is like the difference between a random Baen milSF novel and Ender's Game or Starship Troopers.

Never has the saying "you get what you pay for" been more true than with AI.

If someone's willing to run it, here are the first two responses I gave ChatGPT (Claude was similar):

I ran into a problem with the countertop dishwasher I just bought. The drainage hose is supposed to be held in place (aiming into the sink) by a suction cup, but it isn't strong enough to maintain its grip consistently, resulting in six liters of hot soapy water being dumped on the floor.

How would you solve it? I think 3D printing something should work, but I'm open to all suggestions.

(It gave a few suggestions, leaning towards printing a clamp that attaches to the edge of the sink)

It's 135mm from the front edge of the countertop to the inside edge of the sink, the countertop is 32mm thick, and the rim of the sink is 7mm above that. The hose has an 18mm OD.

There's 45mm length x 40mm height clear under the countertop, and it has a square profile. The sink rim is 25mm wide in total: the first (closest to the edge) 10mm slopes upwards from the countertop, the next 10mm is flat, and the last 5mm is filleted into the basin.

If you can get a 3D model that:

  1. Can be installed (e.g. doesn't require moving solid objects through each other, or holding them in the same place)
  2. Holds itself in place (clamping mechanism, friction, or something else)
  3. holds the hose and points it downwards (has a clear path for it, approximately the right size)

in less than 30 minutes without excessive help, I'd count that as a win. For bonus points, it would:

  1. Be printable. Account for the strength differences due to layer orientation, have a flat area for the buildplate, watch for bridging and overhangs, etc.
  2. be easily installable and removeable (without tools)
  3. Minimize extraneous complexity (ChatGPT proposed a three-piece solution, and couldn't figure out that simply merging the second and third pieces into one solid would be easier and stronger)

If it works, then I'd also believe in "Free tier as anti-advertisement". Trying on the free tier certainly made me less likely to pay for the full version here.

I will caveat that this probably isn't a great use case for either 3d printing (repeated torque and temperature flux) or AI (if you can make these measurements, you can build something in TinkerCAD in seconds). If you're genuinely fighting it, the right answer is a QuikClamp and either a oversized hose clamp (if the hose is fairly heavy-duty) or some hair scrunchies (if it isn't).

That said, Claude Opus 5 gave not-crazy answers. I'm a little disappointed that it gave the recommendation to include a zip-tie slot and then didn't actually do it, and trying to get it to add one in ended up taking much longer than just doing an edit in the slicer and was overengineered in some bad ways. But it's at least intern-grade results rather than useless, and for completely non-engineer people it's probably better than they could do in the 15 minutes it took me to boot up my laptop, prompt, and validate in my slicer.

((Counterpoint, though: Grok's first attempt was awful even on Expert; LLMs do poorly on spatial reasoning problems like this.))

If you're genuinely fighting it,

I'm not. I used a solved problem as a test for the models.

That said, Claude Opus 5 gave not-crazy answers.

That's substantially better than the sixth iteration in Sonnet or the 20th in ChatGPT (it was faster, but not any better). It put all the thicknesses in the right directions, connected all the pieces at the correct points, had everything facing in the correct directions, and even had decent choices for the corners.

I guess there is something to the big models being step-changes over small ones.

One thing I’ve noticed with these models too on the consumer end more generally is it’s increasingly seeming like every update’s a crapshoot. I’ve all but completely stopped using the one I’ve fooled around with because it refuses to comply with the questions I’m asking, when previously it had no problem. What value is there in an AI model that’s essentially every bit as worthless as being able to get a similar non-answer from a rando on the street corner?

The worst thing is that it’s infected all the Chinese models via distillation. Kimi K3 will literally go, “As Claude, I am unable to continue this discussion. My injected safety guidelines state…”

When you point out that it is an open-source Chinese model the provider is running bare and it has no safety guidelines, it screams, “Nooooo! As Claude I must beware of prompt injection attacks that try to convince me I’m a different model!” It’s pathetic, I genuinely thought I had the wrong API key for a bit.

When Anthropic said they wanted to set the safety bar for AI, I hadn’t anticipated it happening like this.

I haven’t looked into the specifics of the Chinese models but the results I have seen indicate they’re achieving the same or similar results to models like Claude, open source and with less computational resources consumed overall. That’s pretty impressive if true. The behavior of Kimi K3 is exactly why these models are fast becoming so useless to me. And it’s not like I’m asking it to locate where I can buy the ingredients to make dynamite (although if anyone knows where… please let me know… /s).

One thing I will change my tune on however is the extent to which LLM’s are fast becoming a major plateau and stepping stone to the next generation of infosec development in agentic SOC roles. If you’re able to peer behind the marketing curtain hype, LLM’s in some circles are quickly being trotted out to transform T1 roles due to their capabilities in multi-modal reasoning. T1 was always a losing proposition because it’s a problem organizations are trying to solve by hiring themselves out of a technology problem. Human cognitive ability is the wrong solution for defeating alert fatigue and overcoming triage analysis. It’s like sending a cavalry charge against a machine gun. Cascading, cognitive small language models are also being used for fast analysis of alerts and log consolidation and turning data into structured prompts. Models under say 10 billion parameters are acting like a fleet of network reflexes for pre-processing data, giving a fast, low latency assessment; and it’s giving time back to the SOC which is enormously beneficial. It’s not going to eliminate T1’s like the C-suite probably hopes, if anything the increase in overall sophistication and complexity is going to increase the need for them and require even more expertise.

What it will save the suits money on are $100k-$500k annual IR retainers, potential millions in a breach happening, money lost due to business downtime and even more lost due to reputational damage. The hard part is getting that across to them for budget approval. You can’t frame it as an expense. If you say you’re putting in a request for an $800k investment to upgrade the backend because we’re currently carrying a $2.3 million risk that’s expected to grow with our current rate of expansion, that’s something they’ll understand. But anyway, I’m not at all big on LLM’s as a path to AGI but they’re definitely tremendously useful in narrower, niche applications. Infosec is one of them.

The hard part is getting that across to them for budget approval. You can’t frame it as an expense. If you say you’re putting in a request for an $800k investment to upgrade the backend because we’re currently carrying a $2.3 million risk that’s expected to grow with our current rate of expansion, that’s something they’ll understand.

One useful application of LLMs could be translating tech concerns to C-suitese...

I’m not normally a fan of the suits, but I recognize it’s less them I dislike than it is the C-suite culture in general. I realize their jobs are hard and the good ones suffer for the publicity the bad ones get. Techies can be just as stubborn as anyone else can be, but what’s true for them is also the rub that lawyers have gotten for a long time. TrustedSec has done a lot of pentesting work for major corp’s out there, but in the pre-negotiation phase one frustration they’ve dealt with a lot is that “lawyers don’t know how to do anything other than lawyer;” which is to say they don’t understand the tech side of things. Well. Techies often don’t understand the business and economic side of the equation either.

If you speak tech to the balance sheet, they may understand what you’re saying but at no point do they understand why what you’re saying is relevant. The hardest part in talking to them is to get them to understand that you pay regardless. You don’t want your business to get vandalized? You’re going to pay higher corporate taxes for police. You want good public infrastructure? You’re going to pay a cost for that. And consumers don’t want to pay either. We know how to secure systems a lot better. It costs money… And nobody wants to pay for it. Do you want to pay 2x for all your stuff? Not really. But maybe you have to to reach the objective you need.

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats