Slightly OT: how fragile are 3D printed things? Do they break easily? Do they feel fragile? The rods in thi design, might a person break them by accident? How about intentionally?
According to the description, the rods are aluminum. Printing rods this long is near impossible and if it was, they definitely wouldn't be rigid enough for this application.
To answer your main question, it depends. There are many materials you can print with, they're strength can differ greatly. Some are very rigid, but somewhat brittle, others are less rigid, but also less brittle.
Then it also depends on the design, the video[0] below is a great video on designing for 3d printing and the considerations needed for strength.
Finally, print settings matter. Temperature can affect layer adhesion. Infill and wall count can affect strength and so can print direction (see video).
3D prints can be pretty strong though. I'm using 3d printed mounts to hang snowboards on my wall, for example. I've printed plenty of other stuff I really couldn't print by hand. And I'm only using FDM printing, with SLA and SLS printing even stronger parts can be printed.
PLA is not strong and doesn't like to warm up in the sun, Silk PLA (made to just look nice on the shelf) can even be brittle. But there are better materials.
PETG is entirely reasonable, I store my pots and pans on PETG hooks and they're not oversized. The superior 3d printers can print ABS which is above any and all doubt.
There are also specialized materials, such as carbon infused PETG, and there are engineering materials at 500€/kg with some interesting properties. But the gist is that PLA makes you go "this sucks, must be printed", and PETG makes you go "oh this is fine".
Then there are resin printers that selectively UV cure a liquid to polymerize it. Last I tried that sticky goo it was awfully brittle and borderline unusuable. I heard it improved, but that's why resin printing is not everyday parts but figurines.
Regular PLA is actually stiffer than PETG/ABS, but it's more brittle (so has less impact strength) and a pure formulation has a lower softening temperature. It's generally not suitable for parts that go under the sun, unless it's some other formulation such as some HT-PLA variant, in which case it's actually a good choice for parts that need to be thin and stiff.
Under most cases, you won't get the same inter-layer adhesion with ABS, so while you get better impact resistance, under most circumstances PETG will yield more durable parts that won't delaminate under the same stress conditions. For outside use under the sun, you should use it's cousin ASA.
To respond to the OP.. 3D printed parts can be incredibly durable when printed correctly. The parts need to be designed for 3d-printing in mind, like most other manufacturing methods. A 1cm-thick 15% filled PLA slab that has been printed vertically might be easy to snap in half with two hands, but it becomes almost impossible to break with bare hands when printed horizontal, and requires a saw to be cut when filled to 50%+. And this using consumer-grade printers.
I'm using 3d-printed parts for work and at home, some in use for almost 7 years at this point, and the only telltale sign is the layered look.
The rods in the design are not 3d-printed, which makes sense (most plastics would be too flexible, and 3d-printing a rod is always more expensive).
> that's why resin printing is not everyday parts but figurines.
It's probably more to do with it being messy, smelly, and overly finicky business[0] compared with FDM (but there's zero chance of burning your house down, I suppose.)
[0] levelling the build plate, cleaning the FEP on failures, replacing the FEP, washing the results with IPA (mostly), having to keep the resin warm to get good results, draining the resin when not used for a while, faffing about with orientation and supports is a black art, etc.
> PLA is not strong and doesn't like to warm up in the sun
"Strong" is ambiguous. PLA is stiffer than many other thermoplastics (resists bending loads well), but it's not as tough (impact strength is poor/it's brittle).
PLA becomes even stiffer and resists higher temperatures than many other thermoplastics if you anneal it.
You want stiffness in a telescope because deflecting under load moves your optics out of alignment. A telescope has moving masses as well, so stiffness along all axes under load is important.
Haven't printed this telescope but I have printed some pretty big things for example a front rack for my wife's bike.
It's on the 3rd version, the first one broke on small collision (the bike fell over while standing), the 2nd on a big collision (my wife hit a bus stop :/ ). The 3rd version survived several years of daily use including hauling stuff from the mall.
The key is to split the design into parts that need to be strong in 2 axis and can be weak in the 3rd axis (to work around the layer adhesion problems). So you print 4 walls laying flat not standing vertically. That plus including rope or zip ties for elastic connections where needed - and you can make stuff that is plenty strong enough for most daily uses.
I printed with PLA and PETG and both are ok - PETG is less stiff but handles direct sun better.
If you've never FDM 3D printed anything before, you'll probably find it about on par with most other plastic objects in your life, even if it is technically weaker (for the most part).
Most things aren't particularly fragile - print out a 1" cube in any material at default settings, and you'll be able to stand on it. You probably won't find a way to break it with your bare hands, short of pitching it into a concrete wall.
The durability depends entirely on print settings (infill percentage, layer height), material choice (PLA, PETG, ABS), and load direction (perpendicular to layers is weakest), but properly printed PETG or ABS parts with 25%+ infill can be surprisingly robust for functional parts like telescope components.
That can be a bit misleading, I feel.
It depends on how well the layers are fused together, post-processing, or even the orientation of the print.
At best, it can be nearly as strong as a water bottle, but I wouldn't expect it to be.
The rods are heatshrink tubing wrapped-aluminum. I personnally have built a few partially 3d printed, or full 3d printed (except for a few hardware items and carbon fiber rods) telescopes – it all depends on print preparation skill, correctly drawing for 3d printing, and wise choice of material and print conditions.
> It may be the wrong one, but I looked it up out of curiosity, so it doesn’t ultimately matter.
Now this has to be the most surprising thing I've read this week, until I thought about it for long.
It doesn't matter if it's correct because you googled it out of curiosity? I think you might be confused. When you're curious, you want the answer to be correct (there's related questions that are raised and consistency checks that a curious person does in their mind.) I think what you are is addicted to the internet. You searched something because that's you brain's kneejerk reaction to whatever the trigger was in this case. The act of searching is what your brain was looking for, not the actual information.
> The authors were aware of many of these shortcomings
Soft sciences are rife with "yeah I'm aware of the problems with the thing I'm doing, but I'll do it anyway. I've presented a disclaimer, that should be enough to cover my ass".
> Soft sciences are rife with "yeah I'm aware of the problems with the thing I'm doing, but I'll do it anyway. I've presented a disclaimer, that should be enough to cover my ass"
It's not to cover one's ass but communicate limitations. If you think the hard sciences don't do this, I've got a cosmic distance ladder to sell you.
> It's not to cover one's ass but communicate limitations.
Ostensibly to communicate limitations; I respect this case. But often times it's to cover one's ass in the guise of communicating limitations.
Hard sciences do it way way less. The reason is that in the hard sciences, using a methodology that "has limitations", depending on what the limitations are, might mean the output is straight up meaningless. Imagine I tell you "I've managed to prove theorem X. Let's start by assuming that 1+1=3. I know it's not, but I'm communicating limitations and let's see where that gets us".
But ok I think we're on the same page, you're just more generous than me.
Think of it like Unicode. The Unicode Consortium’s job isn’t to create character encodings. Instead, it’s to unify encoding that already in common usage. If the encoding that is
In standard usage for a language is missing something, or there’s an issue with it, they’re not going to fix that.
OT does anyone else find it off topic to see the word "grokking"? Does that mean understanding? Do we need a new word for this extremely basic concept?
"Grok (/ˈɡrɒk/) is a neologism coined by the American writer Robert A. Heinlein for his 1961 science fiction novel Stranger in a Strange Land. While the Oxford English Dictionary summarizes the meaning of grok as "to understand intuitively or by empathy, to establish rapport with" and "to empathize or communicate sympathetically (with); also, to experience enjoyment", Heinlein's concept is far more nuanced, with critic Istvan Csicsery-Ronay Jr. observing that "the book's major theme can be seen as an extended definition of the term." The concept of grok garnered significant critical scrutiny in the years after the book's initial publication. The term and aspects of the underlying concept have become part of communities such as computer science. "
It's a pretty common, well-accepted use in the hacker lexicon. See esr's Jargon File [0] where, by some sources [1][2], it started being used in its capacity as meaning 'understanding' for forty-ish years now at this point.
What's your point? It's 60 years old therefore can't possibly be stupid?
Or perhaps you have no point and are just nitpicking that I called it new? Compared to the word "to understand" it's new, it's pretty obvious that my use of the word new had a context attached.
Fellow Branestawm enthusiast here. That is the first time anyone has ever mentioned Professor Branestawm on HN, as far as I can tell! It's triggering deep memories.
> lot of this "rewrite X in Rust" stuff feels like
Indeed. You know the react-angular-vue nevermind is churn? It appears that the trend of people pushing stuff because it benefit their careers is coming to the low level world.
I for one still find it mistifying that Linus torvals let this people into the kernel. Linus, who famous banned c++ from the kernel not because of c++ in itself, but to ban c++ programmer culture.
Reading your comment makes me think that you believe that the point of peer-review is to ensure that a paper is correct, or at least that specific aspects of it are correct. Is that the case? What do you think the point of peer-review is?
I'm not the person you replied to, but I think that in the lay world, people do indeed think that peer review is as you've described. If it's not, then maybe it should be?
Research gets cited constantly in public debates and is used for policy decisions, so the public should be able to quickly separate the good from the bad, the "maybe this is true" from the "this is empirically proven."
The public has lost a lot of trust in Science because research papers have been used to push political agendas, which can then never be questioned because doing so means arguing with a supposed peer-reviewed scientific consensus.
I wish this comment was more representative of my personal experience in science.
Instead I got PIs happy to say that weak evidence "proved" their theory and to try suppress evidence that negatively impacted "fundablity". The most successful scientists I worked with were the ones who always talked like a PR puff piece.
> public has lost a lot of trust in Science because research papers have been used to push political agendas, which can then never be questioned because doing so means arguing with a supposed peer-reviewed scientific consensus
The public has lost trust in science because 10 to 30% of it is scientifically illiterate [1]. (Tens of millions of American adults are literally illiterate [2].)
That's what lets activists and politicians cherrypick bad science that supports their position or cast a scientific consensus as unquestionable.
That's certainly true but I think there's also a very real issue as described by GP. Research gets cited in a political setting as a rhetorical tool. That cycle does a lot to erode trust in the establishment because it incentivizes non-scientists (who are otherwise uninvolved) to behave as though the process is a partisan effort to be interfered with for the benefit of one's "team" rather than an objective pursuit to be funded at arms length for the betterment of society at large.
Obviously reproducing results as part of peer review is not a workable (or even coherent) solution. I don't pretend to have any idea what a solution might be. The obvious issue is that academic publications were never intended as political tools and should not be made into that.
On several occasions I've had interactions with laymen where I found myself thinking "if only you hadn't had access to pubmed and way too much motivation we'd both be better off right now" yet I firmly believe that free and open access to information is a huge net benefit to society on the whole.
> I'm not the person you replied to, but I think that in the lay world, people do indeed think that peer review is as you've described. If it's not, then maybe it should be?
It is not, and it cannot be. It is unrealistic to expect a referee working in their free time to confirm studies that often cost millions of dollars. This is a fundamental misunderstanding of what peer review is and why it is useful in popular or heavy vulgarised science.
Politicians, journalists, and university press offices are guilty of this, and they are those abusing peer review to give some studies more weight than they deserve.
OT I've started referring to the "efficient markets hypothesis" as a religion. I promise you, you can show this article to one of those people and they'll construct a story about how this makes perfect sense.
I did not find that quote in the linked-to interview.
The closest is at https://youtu.be/bM9bYOBuKF4?t=148 "That's the statement of the hypothesis, but it's a model, it's not completely true, no models are completely true. They're approximations to the world. The question is, for what purposes are they a good approximation. As far as I'm concerned, they're a good approximation for almost every purpose."
Humans are too closely related to chimpanzees and Capuchin monkeys to be Homo Economicus. AI will have advantages over humans except that humans still drive the economy and financial activities, and as we've seen with bubbles like real estate and dot-coms, predicting when the stupid ends requires predicting humans.
Suppose you had a bunch of decisions that needed to be made, and you would rather that (on average) they were made by smart people rather than dumb people. So you set up a system of tokens, where people with more tokens were allowed to make more of the decisions.
Then all you need is a system whereby smart people wind up with more of the tokens than dumb people. So you set it up that, when a dumb person meets a smart person, the dumb person gives the smart person some of their tokens.
But how are they supposed to know which is which? "Dunning Kruger" and all that seems to doom this plan from the start. But markets always find an answer; what we're seeing here is the efficient market resolving the question for us.
Awful. All these arguments boil down to "a large number of people couldn't possibly make decisions these bad". Can and do.
To be clear, Im not going to continue arguing this point. I find it as dull as trying to convince a fundamentalist that his arguments are circular. That was fun when I was 14 or so, but it's been a long time since.
To be clear, I was role playing an efficient market fundamentalist. Or trying to.
And (trying to steelman the argument) was careful to not specify which group of people were making the bad decisions. It is obvious that at least one side of any pure monetization trade like this are making a bad decision (and it could be both; for some such games the only winning move is to refuse to play). So I certainly wasn't claiming that people couldn't make decisions this bad since (as you note) this is clearly counterfactual.
My point is that we're seeing an intersection of "you can't cheat an honest man" and "there's a sucker born every minute," with a touch of "we cheart the other guy and pass the savings on to you"; _everyone_ involves thinks they are making a smart move, and in a case like this at least half of them must be wrong.
Your stab was good. Importantly, an efficient market isn't necessarily a rational market. "This is nuts" describes an irrational market, which may or may not be inefficient.
The problem with the reasoning is, either we already know who's a smart decision maker, in which case the mechanism isn't efficient. or else we don't know, in which case the logic is circular - whoever gets the most tokens must be the best decision maker.
in any case i think all smart decision makers have learned to stay away from crypto.
Suppose for simplicity that every sentence in the book is 50 tokens or shorter.
According to the stated methodology, I could give the LLM sentence 1 and have 42% chance of getting sentence 2 recalled. Then I could give it sentence 2 and have 42% chance of getting sentence 3. Therefore, the LLM contains 42% of the book in some sense.
I disagree this is "not really very much". If a person could do this you would undoubtedly conclude that the person read the book.
In fact the number 42% even understates the severity of the matter. Superficially it makes it sound that the LLM only contains less than half of the book. In reality the process I described applies to 100% of the sentences. Additionally I'm guessing that the 58% times where the 50 tokens arent recalled correctly, the outputted token probably have the same meaning as the correct one.
Except it's not what happened, per the article. Instead, they walked down the logits, which is more like asking someone to give 10-20 best guesses for next word, and should one of them match the secret answer, telling them which one is it and asking them to go on with the next word. Seems like a substantially easier task, and most of information is coming from researchers making a choice at every step.
