Member postings for S K

Automatic fact-checking is a top priority for all LLM's, and is in place in most already. Or you can say "fact check that for me, please" and it should provide links, etc.

These systems are impressive and are growing better exponentially. They will change the world, for better and for worse. Prepare to be spammed with auto-generated fake news and believable personalized calls from your "friends" who are in trouble and need you to send money fast. (And, as always, war and porn will be amongst the first and biggest applications.)

I haven't tried one myself yet, partially because of the privacy issue. But also because I don't want to use it for shopping and other stuff Microsoft, et. al., will want you to use it for (it's all about money in the end, isn't it?).

Instead, I'd like to use it to explore the zeitgeist of human knowledge and culture. But one "problem" with the current crop of publicly-accessible generative AI's is that they are deliberately hamstrung. For example, they will refuse to converse on a given topic for more than say a dozen prompts to stop them from meandering. And, of course, they are heavily constrained because of the fear that they will go off into unfortunate directions that may hurt people's sensibilities.

Now we see the likes of Google actually promoting government restrictions on AI. I wonder if that's really just "rent seeking" - the desire to remain the nearly-sole controller and provider of access? Probably, but it's probably also doomed to failure.

There are efforts to develop public domain LLM's (and photo generation, etc.) that will run locally on a reasonably powerful personal computer. For example, Meta's model was made open source, and their "weights" (the trained data) leaked. These are interesting to me, and one day I'll see if I can install one.

The scariest question of all is when they will achieve superior intelligence to us (whatever "intelligence" means), and what they will grow capable of. Don't go thinking that humans will remain on top. At the rate they are improving, it could be just a year or two. And sooner rather than later, they will be handed abilities to interact with the world autonomously. It's inevitable.

Pandora's box has been opened.

Edited By S K on 25/05/2023 16:05:10

A little more progress: I've completed the stage, which takes the place of the plastic version seen above. It's made from 510 bronze, with a brass ball on which the micrometer pushes. I gingerly tested the flexibility of the hinge, and it will move, but I may have to thin it a little more - not sure yet.

I've decided to fit the "slides" (10mm glass windows) into little 3-D printed holders, which then drop into the 12mm hole. In the event that I get the lens-to-slide distance badly wrong, I can reprint the holders to adjust the slide's position.

Still left to do is create a few mm thick shim with a very slight angle to set the distance between the body of the microscope and the stage, and drill the microscope's body to accept the stage. However, I'll soon need to mount the lens, at least temporarily, along with a slide, in order to judge how thick the shim needs to be, and perhaps what kind of angle to set. Then I'll be nearly done.

Edited By S K on 20/05/2023 21:03:08

Oh, yes, I've thought about interchangeable lenses! The lenses are so small that they could easily be mounted in threaded receivers, to be screwed into the main body. It would be easy since I already have the materials and small taps and dies that would be needed.

But that might be gilding the lily? I am happy with it so far. Maybe for the Mk. II version. 🙂

Edited By S K on 19/05/2023 02:33:45

Not really. I wanted to maintain at least some resemblance to the original, and a steady flat face is a virtual necessity due to the shallow eye relief. He got that part right.

I ordered a sheet of bronze this morning, and it came this afternoon. McMaster is amazing.

It's cold-worked 510 bronze, and it definitely has a lively, springy feel compared to the dead feeling of brass.

Bronze ... for when brass just isn't expensive enough! 😄

I'll order some. 🙂

Yes, it's similar to a bell crank in function. Haha, haven't used the term "bell crank" since I was a kid making control-line model airplanes.

The effective focal length of the lens is 4.5mm. But because the lens is 3mm thick, the back focal length is only 3mm (about 4.5mm - 3mm/2). So a flexure-supported table should be about 3mm away from the base. I had imagined using 3 or 4 of the flexures I printed (or something like them - it was just a generic test structure) to support a stage at its edges. But it would have to be scaled down a lot further still, which looked to be a big problem - the arms were already only two extrusions wide. Maybe photo-etching could make parts small enough.

Anyway, the hinge it will be.

Referring to the hinge area of the printed part, I wonder how thin I have to make brass before it will bend appropriately over a few degrees? Or is it going to fatigue quickly and just break? I suppose I should test that before going much further.

Edited By S K on 18/05/2023 13:10:24

Most printers should have a profile for PETG. I have a Prusa, and the usual profile runs at 240C for the nozzle and 85C for the bed.

I use their textured coated spring-steel bed plate. It's awesome. Zero problems with adhesion, easy to pop the prints off, and it still looks like new after extended heavy use.

I printed out a hinged stage to see how well that might work, and it works pretty well. It's just clamped in place so I could test its behavior.

It's a simple technique, though it has the disadvantage of losing parallelism to the lens while focusing (the original had the same problem).

The translation ratio (stage movement / micrometer movement) is a little below 1:1 at the lens, so a small reduction, which is fine - I wanted to avoid amplifying the micrometer's motion.

I think the right strategy would be to position the stage tilted slightly up (higher) than the nominal focus point when no micrometer pressure is applied, but such that the stage is flat when the micrometer moves it to the nominal focus point.

One of the goals of this project is to gain some experience with machining, so I'd like to do this in brass, too. But the prints are very helpful to test ideas and work things out, including the important task of calibrating the point of focus.

Edited By S K on 17/05/2023 22:16:49

Between PETG and PLA, go with PETG for sure. It's seriously tough stuff!

It's easy to print on almost any printer, and I use it for almost everything. Anything much tougher than that will likely require higher than typical temperatures and an enclosure for the printer.

Edited By S K on 17/05/2023 17:56:13

I printed a primitive flexure as a test. It wasn't printed to proper scale or with any real thought.

It certainly is flexy, but I'm not very happy with how it moves. It's actually too flexible and I think it would have to be stiffened a fair amount. It's printed in PETG, which I use for pretty much everything. Maybe PLA would be better.

It's also becoming clear that miniaturizing an entire table that moves properly on flexures will not be easy. I might abandon this approach and go with a much simpler hinge-type mount. That has the disadvantage of tilting forward or backward out of the plane of the lens as it's focused (as Leeuwenhouk's did), but it may suffice.

I also ordered a few 10mm diameter, 1.1mm thick borosilicate glass windows (uncoated) to use as subject slides. When I was a kid, I built crazy things out of bent nails and scraps of wood, and proper materials were simply unavailable. It's so awesome to now be able to order exotic materials with a click of a button!

Edited By S K on 17/05/2023 16:38:32

I've made some more progress. 🙂

I had a small 1/2" micrometer head in my drawer, and thought it would be a perfect addition. I'd like the micrometer to perform focusing rather than vertical translation as in the original. It may be sacrilege, but I think it works! 😀

The front, facing towards your eye:

And the back, facing away:

I will explore some sort of flexure for focusing, but I think I'll do that in 3-D printed form, at least at first.

Fun! 😄

All words start out as meaningless nonce words. Then they gain meaning by being used in relation to other words. For example, I can't understand what teenagers are talking about because of all their secret new words that haven't seeped into general use yet.

So yes, of course the word "stop" has meaning, but that's only because of its prior and now established connections to other words, for example the words "here's your ticket for blowing past that stop sign!"

So, it's the network and strength of connections between words that are their meanings.

Edited By S K on 16/05/2023 21:01:36

I invent a new word: "blarging."

What does it mean? Absolutely nothing!

Now I put the word in relation to others: "I blarging hate that blarging stuff!"

And presto, now it has some meaning! 😄

Words only have meaning as a network of connections amongst other words.

In the old days, AI was created by humans teaching computers how to do things such as play chess, e.g., taught by humans that a pawn is worth 1 point vs. a queen worth 9, and given a hand-assembled database of opening positions, etc. After that, they just took advantage of the ability of computers to test millions of combinations. There was little in these systems that people couldn't do or understand in principle, because they taught it in the first place.

The next phase (e.g. in chess) was teaching strictly only the rules, e.g. only how the pieces move, and nothing at all about how to play well, but then allowing them to learn how to play chess well by playing millions of games with itself.

Similarly, the latest "generative" AI isn't explicitly taught anything by people. It's just shown millions of documents and it learns on its own that, for example, the word "Merry" is frequently followed by "Christmas," and so on. So then when it's given the prompt "please write a poem about Christmas," it just starts to add one word after another based on the statistical strength of connections between words. To the shock, delight and horror of people, if you allow it to form enough connections between words (we're talking billions), its statistics-based production starts to make rather good sense!

Generative artwork is a little different because it's 2D vs. linear like text. Very roughly, it works like this: You show a computer millions of cat photos scraped from the internet, and (via magic neural net stuff) the computer learns what a cat looks like. Then you give a computer just an image of pure noise as a starting position, but you tell it "that's a picture of a cat, now please get rid of the noise." So it starts incrementally deciding "this little group of random noisy pixels looks just a tiny bit like it could be an eye, and if so, those over there could be the other eye..." So it makes a small refinement that steps the noise just a tiny bit towards a cat. After dozens or hundreds of steps, an original picture of a cat emerges from pure noise.

Now everyone is wondering "so computers can think now? No! Surely it's just following one word after another not from knowledge but from statistics! That's not thinking!" But the deeper question is "isn't that what we are all doing too, at least most of the time!?" Words don't have meanings in isolation. The information is in the connections between words, and now both computers and us know those connections. That's why I don't really have to stop and think about every individual word as I'm typing. They flow as they do because I know automatically from past experience reading thousands of books and documents that this next word follows all those that I wrote before, just as computers now do!

This is going to be a wild ride!

I've been thinking about the focusing mechanism. I'm not trying to make a strict replica. Rather, I'm trying to incorporate a few improvements while maintaining the readily-identifiable appearance of being a "Leeuwenhoek microscope."

For one, I've been intending to use a small round piece of glass as a slide on which to mount a subject rather than the needle that Leeuwenhoek used. This would require some sort of holder that I haven't fully figured out yet.

On to the focusing mechanism: Leeuwenhoek used a screw (the focusing screw that sticks outward from the back) that applied a force to pry the entire assembly apart, thereby adjusting the distance from the lens to the subject. This has several issues and side effects that I'd like to improve on. For one, this bending changes the plane on which the subject is in, tilting it forward or backward relative to the lens. It also slightly shifts and rotates the subject. And then of course there's the bending itself, which I'd prefer to avoid entirely if possible.

I can imagine a few ways to focus while maintaining parallelism of the slide, but so far my ideas tend to be either complicated or difficult to miniaturize.

Any brilliant ideas from the readers?

Hehe, it would be better if I figured it all out in CAD before even starting, right? But too late for that now! Onward! 🙂

Edited By S K on 16/05/2023 15:23:03

A superb site for information on Leeuwenhoek and his microscopes:

lensonleeuwenhoek.net

And here is a research paper in the journal Science in which neutron tomography was used to analyze the lenses in two of his microscopes:

Tomography paper

One of the two lenses - the lower power of the two - is "lentil" shaped (more or less like a standard bi-convex lens), and one is a sphere. The spherical one also has a little stem, presumably left over from the lens-making process, e.g. a stem on the end of which a spherical drop was formed.

In one of his letters, Leeuwenhoek stated that he ground his lenses, and at other times he derided spherical ones (he was in competition with Hooke at the time, who used them), but apparently he used them too in at least some of his microscopes.

Edited By S K on 14/05/2023 18:31:41

Thanks for providing links to originals and other replicas. One link mentioned how notoriously difficult his microscopes were to use, and I'm finding that myself, too!

High-quality achromats like the one I'm using are for general scientific use, with no one special purpose in mind. But if you have seen pictures of optical tables with lasers shooting around everywhere, that's one application for them. They are available in a very wide range of diameters and focal lengths. You can also get them with other coatings for better transmission in the ultra-violet, and so on (magnesium flouride coatings are only good for visual light). Very tiny lenses like this one are more expensive than average. This was bought from Edmund Optics and cost a little over $100, while an uncoated single-element lens could be had for pennies.

An interesting point is that, back in Leeuwenhoek's day, his single high-powered lens approach (essentially an extremely powerful magnifying glass) was more successful than compound microscopes were. A compound microscope includes an objective and an eyepiece, which when combined multiplies the power (total power = objective power x eyepiece power). However, optical aberrations were pretty bad back then, and in compound form the net result of multiplying those aberrations was simply worse than the best single lens. It wasn't until the invention of achromatic lenses (like I'm using here) that compound microscopes became practical.

The periphery of Leeuwenhoek's lenses were blocked from use by the frame they were mounted in. Whether he understood this or not, that's equivalent to "stopping down" - constricting the aperture of the lens. Aberrations from the periphery of a lens are typically far worse than in the center, and stopping down blocks the worst of that. But my lens is quite high quality and should be good out to its edge, and so I wanted to leave the entire lens surface clear, hence the reamed hole. The fit I have is a close sliding one, with just a tiny bit of binding. But I hadn't considered thermal expansion. I guess I'll just take my chances with that.

I've started a Leeuwenhoek microscope project. There have been many replicas made by others, but this one will be mine. 🙂

Leeuwenhoek did not exactly invent the microscope, but he is famous for the many original findings made via his instruments, enabled by his secret method of making superior lenses - a secret he took to the grave.

His instruments contained a single lens element of high power. They were originally thought to be tiny glass beads (spherical lenses), but some recent examinations suggests some or all were bi-convex in shape.

Instead of blowing or grinding a lens, I simply purchased one. It's a tiny 2-element optically-coated achromatic lens that is 3mm in diameter by 3mm thick, and with a 4.5mm focal length. This makes it very high powered at 222 diopters (a typical magnifying glass is 4 diopters), and is in the neighborhood of Leeuwenhoek's various iterations.

An achromatic lens is typically made from two different glasses (classically, "crown" and "flint" glass), usually cemented together, such that the two glasses partially cancel out chromatic aberration. This lens also has a magnesium-flouride anti-reflective coating. (It was not cheap!)

Using a 2-element coated achromatic lens is an unfair advantage, since achromats and optical coatings were hundreds of years into the future for Leeuwenhoek. But if he had these technologies available, he certainly would have used them!

Here's a picture of the lens. You can see where the two elements join.

I temporarily mounted it and managed to view things through it, and I have to say it's quite difficult to use, especially due to the very shallow eye relief. In retrospect, perhaps I should have bought a larger-diameter and longer focal length lens. But once you get it right, it provides a surprisingly high quality and highly-magnified image that I'm sure Leeuwenhoek would have enjoyed.

Leeuwenhoek mounted his lens sandwiched between two thin sheets of metal that were then riveted together. I chose to use a solid piece of brass, into which I drilled and reamed a 3mm hole as an exact fit for the lens. This avoids obscuring the lens' periphery. I'm a bit terrified with the idea of using a retaining adhesive on it, though!

As the brass was slightly thicker than the lens, I added a countersink on one side so the lens can sit a tiny bit proud on both sides. That way, the lens can hopefully be kept clean a little easier.

The body of the microscope is paddle shaped, about an inch by two inches. It fits comfortably pressed against your cheek and nose. This is essential, since with such a shallow eye relief you need to brace it steadily lest you poke your eye.

Here's the body in its final shape, with the reamed and countersunk hole for the lens. The lens is sitting on it.

So just a little progress so far in a relatively simple project. Still to be done are to add a translation and focusing mechanism, a clip of some sort for holding the subject, and a handle.

His microscopes used a point to hold the subject. Instead, I'd like to make a holder for microscope cover-slips. I'd also like to improve on his focusing and translation mechanism while retaining the spirit of his approach. If I fail at that, I'll just do it Leeuwenhoek's original way. After all, he reportedly made well over 400 microscopes(!), so he likely went through many ideas.

Thanks for reading.

Edited By S K on 14/05/2023 02:25:56

• Relativistic space-time effects. There is no such thing as "time" as a stand-alone feature of the universe. There is only unified space-time. My "time" and yours are necessarily different, and it doesn't even make complete sense to speak of time as a discrete thing. So what does your pendulum even measure, really? 😛🙃