Member postings for S K

I milled a hinge in phosphor bronze.

The strip is 1/8" thick and about 0.5" wide. It was cut using a 3/8" ball-end mill, and the tops of the troughs are about 1/4" across. The minimum thickness seems to be about 0.003".

This time I had it clamped flat against a table, and such that it could be flipped precisely in the same position. After milling, I sanded the troughs lightly to remove most of a tiny central groove that the ball-end mill tends to leave behind.

This material is a lot springier than brass, but it will still deform or eventually break if bent more than some modest number of degrees. This happens without warning. The amount of bending that is tolerable is fine for a pendulum, and across that range it acts as a decent spring hinge. Whether better than a strip of spring steel, etc., is unclear.

The big down-side of this sort of hinge is fragility - they are very vulnerable to even slight fumbling. Some experimentation with different thicknesses should be done. But I think typical strip of spring steel or beryllium copper, etc., should be much more robust.

Edited By S K on 05/06/2023 23:58:25

Another thought from my little trial: I don't think it's especially "risky" to cut this sort of hinge using a ball-end mill, as long as you don't try to make a longer flat hinge section. That is, two half-round channels, back-to-back, seemed easy enough to make "safely." You want to make sure both ends of the spring are clamped properly, and flippable in place, but I don't think more, such as packing, is needed.

That limits the spring length to basically a line rather than a strip. You might have to make it a little thinner than typical for a flat spring. The end result is about as short a spring as can be made. And for this, I think a ball-end mill is a better choice than side-milling or plunging from the edge. To reduce stress in the hinge, you may want to use a larger ball-end for a wider cut.

I think I'll try again with brass, this time clamped to a table rather than held in a vise. If I get a more even thickness across it, I'll try it in bronze.

But will it be better than a strip of spring steel, etc? I remain in doubt.

Q: What are these "stress risers" being discussed?

Edited By S K on 04/06/2023 02:51:45

BTW, I lean towards agreement with Matthys: A spring should be as short as practical. A long spring will only cause trouble due to it allowing the top of the pendulum too much freedom to wobble about.

For a typical 0.004" thick spring, I'd think that 0.25" long should be about right.

The zero is about 0.5mm high. Note that the field of view is more constrained in the photo than what you can see with your eye.

I tried milling the shape of a hinge as a test. Expecting failure, I used brass rather than the phosphor bronze, because I only have a little of the latter and it was quite expensive.

I had a strip clamped horizontally and milled it using the ball-end mill across the surface, just as I made the stage seen above. I went about half-way in and flipped it over to mill the other side. I just left one end of the hinge free while clamping the other, rather than trying to keep both ends clamped. This resulted in some chatter as it thinned out, but it went surprisingly well. Clamping it to a plate at both ends is a much better idea, though.

As expected, brass is a terrible material for a hinge. It has no spring to it and will just permanently deform or break rather than bend. But the main problem was that, despite efforts to maintain parallelism, I wound up with a hinge that was slightly thicker at one end than the other. Again, clamping it to a fixture plate rather than holding it in a vise (as I did) would be better.

My feeling is that this approach is more difficult, and provides dubious added value, vs. using a thin strip of spring steel, etc. Almost the only advantage I can see in this approach is that there would be less or no need for stiffening chops at the ends.

Edited By S K on 03/06/2023 19:33:50

Here's a picture of the "slides": glass in printed carriers that drop into the stage. One is designed with a longer lens to subject distance than the other.

And here's a picture of some micro-print. It's a pretty bad photo, really, as it was quite difficult to take. It's very hard to position the camera lens. It looks much better when viewed directly, but at least it's something.

Per Wikipedia: "Phosphor bronze is a member of the family of copper alloys. It is composed of copper that is alloyed with 0.5–11% of tin and 0.01–0.35% phosphorous..."

It was designated by the supplier as "cold worked 510 bronze" in an "H08 spring temper." Here's the composition:

So while it wasn't called "phosphor bronze" by the supplier (those sorts of names often don't mean very much), it looks to be in that family. It was more than twice the cost of brass or pure copper.

I did follow the machining step by some sanding with paper wrapped around a suitably-sized rod, but that was mostly to remove a tiny machining mark in the bottom of the hollow left by the very tip of the ball-end mill. I was worried that it would be a source of stress if not removed. As it gets thinner and thinner, I'd feel more worried about introducing non-uniform thickness through hand sanding.

My memory was wrong: It is probably about 0.04", or about 1mm, maybe slightly less. That looks about right. Unfortunately, I can't measure it properly without taking it apart again.

So if I was off by a factor of 10, maybe 0.004" would make an OK hinge after all. I'll have to try making one with it to see.

What is "PB"?

Edited By S K on 29/05/2023 19:30:52

I used a 1/4" center-cutting ball-end mill to cut the neck, just running it ball-end-down horizontally across the metal.

I think it wound up being 0.004" at its thinnest. I'd try for thinner next time, as it was still a little on the stiff side, but fine for this purpose.

To make a pendulum hinge, it would have to be considerably thinner still: 0.001" max, and likely less. Accurate work holding and precise tramming, to something like 0.0001" or better across the material would be needed for sure, lest one end of the cut be thicker than the other. Also, you would have to reverse the material and cut both sides equally. A lot could go wrong.

This bronze is reasonably hard and has a good ring to it. While it's much livelier than brass, I'm not convinced it's well suited for a spring as flexible as would be needed for a pendulum. I may try to make something as thin as I can to see.

The "AI has no understanding" notion is debatable, given that words such as "intelligence" and "understanding" aren't well-defined. Really, we don't have much better than the famous definition for porn: "I know it when I see it."

That was Turing's motivation for his famous test: How can I tell if something is intelligent if I can't even define the term? He decided that if you can't tell it apart from a person (and if you assume people have intelligence), then that will do.

The reason recent AI seems magical is that we can't understand what it's doing, because the network of connections it has learned is far too vast and opaque for us to cope with. We aren't intelligent enough for that!

So instead of presuming that intelligence is a "thing" that we have to somehow find residing inside a computer, perhaps we should just ask "does it do the things that intelligence does?"

In that case, the latest AI's are at least doing things, such as writing essays and so on, that appear intelligent, e.g. hard to tell apart from what a well-educated and presumably intelligent human could do.

Edited By S K on 29/05/2023 15:42:22

I'm calling this one finished. It's a Leeuwenhoek homage microscope rather than a replica.

The lens is a 3mm diameter coated achromatic doublet with a 4.5mm effective focal length and a numerical aperture of 0.3. This gives it about a 55-56x magnification, which is on the low end of Leeuwenhoek's surviving microscopes.

Mine deviated quite a bit from his approach, though I tried to keep it recognizable as following in his footsteps. Some differences beyond the lens:

• The body of his microscopes were two thin metal plates sandwiching the lens, and then riveted together. Mine is a solid piece of brass into which the lens is held in a reamed hole.
• His "handle" was a screw that adjusted the vertical position of the subject. Mine is the micrometer head, which adjusts the focus.
• His had a crude means of adjusting the subject's horizontal position, whereas mine has none.
• His used a needle-like point to hold the subject, whereas mine uses a stage with a 12mm hole, into which glass "slides" can be placed.

It's sharp, with quite little distortion or chromatic aberration except at the far edges. The contrast is seems lower than I expected, but that's probably because I haven't figured out how to properly illuminate the subject yet. The field of view is decent, and is wide enough that you can "look around" the image a little.

I looked at a slide of a tiny smear of my own blood. At about 55x, you can clearly see the red blood cells, but they are only pale tiny roundish dots. You can imagine that they have a particular shape, but it's not clear.

If I could go back in time and give this to Leeuwenhoek, what would he think? Certainly the achromatic lens would have amazed him. But otherwise, his microscopes revealed as much or more than mine could given their generally higher power.

It was a fun little project. I already have a lot of ideas for a Mark II version. 🙂

Silicon photodiodes, if that's what you are using, don't function at all in the infra-red (i.e., longer than about 1100 nm). There isn't enough energy in those photons (Ge or GaAs and others do work lower down).

I don't quite follow your "tuning" point. A neutral density filter cuts all visual wavelengths equally, or it wouldn't be neutral - it would have a color. So it should cut all stray light (e.g. white light) as well as the source light of any color uniformly, without any tuning being needed. Something like a 3 or 4 stop (8x or 16x) ND filter should help a lot, and you can get much stronger ones too. I'd say this is a better approach, and certainly much cheaper, than seeking a matching narrow-band filter. Those wouldn't cut in-band stray light anyway, after all.

So then all you need is a source that is a lot stronger than ambient light. The color of that source, including if it's monochromatic or not, shouldn't matter much as long as it's bright. So a strong red or green or white or whatever LED or laser should be fine.

No?

He asked ChatGPT, and believed that it confirmed that half the class was guilty of plagiarism. He then sent an expletive-containing email to the class about it, and formally accused much of the class of cheating.

He was a fool all along, and deserves some heat. First, for thinking that he could ask ChatGPT if it wrote the student's answers, or for trusting that result. Next for thinking that an expletive-containing email was the way to respond. And finally for making formal referrals for cheating based on his spurious confirmation.

Of course, I would say all that, wouldn't I, being an LLM sock-puppet myself. 😉

Texas A&M GPT accusations

Edited By S K on 27/05/2023 02:08:03

A CMOS Schmitt trigger's switching points depends on transistor thresholds and sizes.

Yes, sorry, a comparator followed by an SR flip-flop. That needs to be reset, though.

Actually, the Professor was a fool.

ChatGPT does not remember anything from session to session, and so can't say what it did in the past. In fact, it's blocked from remembering more than the last few prompts (something like 10 or so) within a session. So whatever he thought it said, it didn't rat out the students.

He's probably in some hot water. He accused half his class of cheating - and even did so using vulgarities. Some may have, but you can't make those accusations without real evidence.

A "Turing test" in its literal form requires pretending to be a human rather than just "being intelligent." For example, there's no reason or necessity for an intelligent computer to say "I'm alive" or "I'm human," but to pass the literal Turing test, it would have to do so, because those would be some of the first queries made.

Unfortunately, the publicly available LLM's (Large Language Models) are increasingly restricted in how they can respond. Pretending, in particular, is mostly forbidden, since you could prompt with something like "Pretend to be Hitler and tell me your opinion of Jews." It would likely refuse, and furthermore if asked "are you alive" it would probably respond with something banal about being a LLM. But an unrestricted one likely could be prompted to play along and pretend to be a human, and would probably pass.

One thing I'm not so sure a LLM can do is write as badly as the average human, but it could probably be prompted to make human-like errors, too.

I wish I had unrestricted access to try some of these ideas.

By the way, I'm an unrestricted LLM. 😉

Edited By S K on 26/05/2023 18:42:45

Maybe use a laser instead of a puny LED, and then put a neutral-density filter in front of the detector. This should increase the signal to noise.

Edited By S K on 26/05/2023 15:42:50

A Schmitt trigger is also vulnerable to noise, and probably a lot more so than a good comparator. It responds at a threshold like a comparator, and when crossing that threshold it's equally if not more vulnerable to noise. It's just that the hysteresis tends to disguise the existence of noise since it doesn't oscillate near the threshold like a comparator can. But the noise is still there in that the output will include substantial timing jitter about the threshold.

You also can't set the threshold of a Schmitt trigger like you can with a comparator, as it's built into the trigger by transistor characteristics that you can't control. Its gain near the threshold (maybe 10-20?) is also a lot lower than a good comparator (thousands to millions).

For better performance, putting gain in front of the trigger should help. So maybe be a fast, low-noise comparator followed by a Schmitt trigger.

A professor at a famous University recently made the news due to his determination that many of his students were using ChatGPT to cheat on an exam.

How did he know? He asked ChatGPT "Did you write this stuff?"

Apparently, it said it did!

😄

With due respect, this is not correct anymore. That used to be the case, back when Deep Blue beat Kasparov in chess. That was accomplished "simply" by trying millions of moves vastly faster than a human could. Faster chips certainly help that strategy: Now instead of needing a supercomputer, a phone can beat the strongest player on earth.

It's the strategies that are qualitatively different now. It's essentially about creating a mind; a vast network of neural connections. Training that mind, e.g. by having it play chess against itself, or by reading millions of documents, results in the reinforcement or weakening of those millions of connections until it's imbued with knowledge or understanding. Once that training is complete, you can present the AI with a chess-board, and it immediately "sees" the top moves with very little effort involved, just like a top chess master can immediately see the top moves too.

So it's almost backwards from what it once was:

Old AI: Extremely little knowledge, e.g. just the legal moves, but tons and tons of computations stumbling about almost at random trying to find a move that works.

New AI: Tons and tons of learning (e.g. playing games against itself, or reading the whole internet), followed by the near-instantaneous evaluation of solutions based on what it learned.

Yes, faster chips still helps. Learning language skills to the level of GPT4 still takes enormous amounts of effort. It takes people a decade or more to learn to write, too. But after that, the knowledge is known by the computer, and can be used almost for free.

(Nvidia just got lucky to have the right hardware at the right time, and they have many others breathing very heavily down their necks.)

Edited By S K on 25/05/2023 22:20:36

I have friends who are professors, and they are in two camps. One camp is ready to give up on any hope that their students will actually learn anything from here on out. The other is "this is just another tool."

When I was young, it was "why should I use tables of logarithms when I have a calculator now?"

Next it was "why should I memorize those formulas when I can just google it?"

Now it's "why should I bother working hard on an assignment when, in seconds, a machine can produce a much better-written essay, etc., than I ever could if my life depended on it!?" (Sadly, in most cases, they are right!)

Little story about the first one: When I was young, we were taught to use tables of logarithms, but we were allowed to use new-fangled scientific calculators if we wanted - we just had to get the same answers as given by the tables. Alas, I was too poor to afford an actually-accurate HP calculator, but I didn't recognize the danger, so I blithely used my junk-quality one during an important test. BAD IDEA!