Recent Posts

1

AI / Re: AI destroying economic system

« Last post by Siwastaja on Today at 12:22:38 pm »

AI datacenters become overloaded with customers as they cannot maintain the ever increasing demand due to a shortage of low-nm silicon.

Price per token increases 10x and projects that were formerly viable become non-viable.

I was taking these seriously 6 months ago when the Claude API prices were high, Claude was slow, "rate limited" API errors occurred constantly (besides the actual 5-hour/weekly limits I never hit).

Now, while more and more users have appeared and AI usage has increased, not decreased, Claude API price has went down, it runs faster than ever, and has fewer outages than ever before.

I take this massive speed and smoothness as a proof that the resource spend has really gone down. (Price cut alone isn't a reliable proof because that's marketing; they could do it on loss, but I don't think they do.)

Cost regulates the usage; it does not prevent it. It's the same as coffee. Finland is #1 or #2 coffee consumer in the world, but a lot of the coffee went down the drain. Just make a liter of it, drink half of it, put the rest down the drain. When coffee price doubled or tripled (!), people just stopped pouring it down the drain.

Same with AI. I never understood what people actually do when they talk about spending their whole MAX20x plan weekly usage limits in a day, when I can't spend MAX5x usage limits.

If the question is about surviving and maintaining working software, the situation is totally different from the "oh fuck we have this new AI thing, crazy, let's try forking farm of 20 agents to work on this poorly thought out problem and then toss the results" madness that was all the rage in March 2026.

So I think the scalability / cost increase problem is a self-regulating. We get truly important work done, so we are not in any sort of crisis because of that. It might even be advantageous if it's a bit more costly and requires a bit more human regulation to choose what tasks we can afford doing.

That leaves your Taiwan invasion / TSMC scenario. It's basically the same scenario but on steroids: with no new manufacturing, the demand-supply curve is even harsher, so price increases / usage limits will be even stronger.

But clearly the current capacity serves current customers just fine. The core question is, how many new customers there will be? How much new software development capacity humanity still needs, and how expensive can it become until we regulate our software output?

2

Test Equipment / Re: Keysight 6.5 Digit Bench DMMs: 34460A, 34461A, and 34465A

« Last post by 2N3055 on Today at 12:14:33 pm »

So, who's using AI to make new firmware for these meters? I know somebody must be. 

Yeah loading some Terminator hallucinations to 1800€ meter is all we need. And then please use that meter for production of nuclear power plants and airplanes..

3

Jobs / Re: mini FDIS Prod Dev - End to End small project - EOI

« Last post by nctnico on Today at 12:14:24 pm »

I assume this project has not been filled yet? The problem with this project is that it needs an engineer with experience designing these kind of distribution amplifiers and access to a good phase noise analyser to asses whether the design matches the requirement. That is not going to be cheap. I'd drop the Kicad requirement for starters so you are also fishing in the pond where the professionals are in (who typically don't use Kicad).

4

General Technical Chat / A Parable for the Times

« Last post by watchmaker on Today at 12:11:46 pm »

A maiden was walking  home one evening in sleet and rain.  She discovered a viper out in the cold, it had been injured by a passing cart and had been unable to get to its den.  Saying "You poor thing, I will take care of you", she picked up the snake and kept it close to her bosom to keep it warm.  At her hut, she made a nest for it near the fireplace to keep it warm and she nursed its injuries.  After the snake began to recover, she picked it up and sat with it on her chair near the fire.  While she stroked it, the snake bit her in her neck.  She exclaimed "Snake, WHY did you bite me?  You know I will die.  I had taken mercy upon you."  As the viper slithered out of her dying hands, it replied "Silly maiden, you knew what I was when you picked me up".

This is as old as the Greeks, it is a variant of one of Aesop's fables.

History is important for predicting how human communities will respond to similar situations even though separated by centuries.  In a way, Isaac Asimov was write (pun) on.  Or as Kurt Vonnegut would say "So it goes".

5

Manufacturing & Assembly / Are there any users here who have Bungard CCD machines?

« Last post by juanenrac on Today at 12:09:14 pm »

Hello, I recently bought a 2003 Bungard CCD MTC with the old MS-DOS control unit.
I am looking for users who own the same MTC model or have the ATC option installed. If you have this machine, please let me know your model (MTC or ATC), (RS232 or USB connection) and its production year.
I am currently developing two separate cross-platform software programs for this machine using its original control unit (RS232): one for PCB prototyping and another for 3D CNC work. I am looking for testers to try them out and provide feedback. Thanks!

6

General Technical Chat / Re: Flying Tesla Roadster

« Last post by nctnico on Today at 12:04:17 pm »

2) Large range (500km EV's are nothing, ICE can do 800km and the millage gets better the faster you drive unlike EV's which is a opposite)

The real winner there is hybrid, with the latest BYD Qin L able to go a staggering 2100km before having to refuel! https://insideevs.com/news/721577/byd-qin-test-range-china/

It is no wonder. With a 60 litre gas tank and very low speed limits (right at the sweet spot between engine efficiency versus drag) a hybrid can get very far.

7

AI / Re: AI destroying economic system

« Last post by paulca on Today at 12:01:01 pm »

We both know there is middle ground we agree on. 

What I see that I don't like is features driving the usage of a component and then that component generates drive for features.  Those features now get rolled on into the next project as the "Base template" if one is used.

Im sure everyone who played with Arduino or ESP32 has done this.  You start with a basic hand written "Wait for Wifi".  Then you add the awkward one you didn't see coming, "Wifi disconnect and reconnect" handling.  Then someone says, "Why don't you use X Library?"

You try it.  It solves your problems, takes care of the corner cases, some you hadn't even considered at that stage..... but it also adds functionality and features your original spec did not have.  You have a "Fall back AP" mode.  You can support better auth methods.  You can persist the credentials into the EPROM or flash instead of hard coding them into the repo.

Now all your projects come with those features in their spec.

In enterprise world of Java the master of this "pull in" is Spring framework.  That list I provided of all the things an application needs.... spring has a component library for it.  They all work together under "Convention over configuration".  Thats a polite fancy sounding way of saying "It has EXTREMELY strong opinions."

Spring starts to spiral.  "Since we imported this, we have access to that, so lets use that."

The other thing you hint to, I agree with, but it's not an engineering problem.  Its a management and business model problem.

I am literally working on something right now, that audit logging thing.  Im starting to struggle and not with the terchnicalities or the engineering...

But... if my team lead answers one more question that contains a "Maybe" or a "Probably" or a "Possibly" I'm going to lose my temper and start shouting.

"The code doesn't have a fucking maybe or a possibly and if you want all possibilities you can wait a few years.  I NEED REQUIREMENTS.  I can do a million different things here.  Only one of them is what the business want.  DECIDE.  CHOOSE.  Tell me what you fucking want!"

But... I am trying to learn that in a consultancy, my model says, "STOP!  Do not proceed until you know where you are going and what you want.  Where you are and where you are going = bearing = direciton = plan = close".

In the modern consultancy world, the model is.... "Make it up", "Burn some days on it", "Lets talk about it.", "Show us something"... then they tell you its wrong and make some completely bizarre non-technical requirement up on the spot which fails it outright and it needs scrapped and redone... but you are not given the time to do so, you are expected to hack it to make it work.  You complain that will leave technical debt and a manager literally laughs in your face and dismisses the concept of "Technical debt" as something made up so Engineer could spend more time chatting at the coffee machine or gaming.

Its infuriating.

8

EEVblog Specific / Re: EEVblog 1784 - Claude Code Reverse Engineer 121GW Firmware SUCCESS!

« Last post by VinzC on Today at 11:55:30 am »

If a program is not pure C, it cannot always be translated to C. Imagine its using a naked ISR function or an inline assembler section thats jumping around, etc.
Or you could even jump midway into another "function" as long as the stack/registers are managed the same way. Is this now 1 C function with 2 entry points? Doesn't make any sense within the mindset of a C program!

For having inspected a lot what code C and C++ compilers throw onto my hard drive, I can tell that even compiled C functions do not always look like they were C anyway. Jumping into a function at a different address is common with compiler optimisations. I've seen that a lot with GCC. I don't know Ghidra at all but I would expect it knows optimisation patterns in order to reconstruct potential candidates that match the resulting assembly.

EDIT: And as an example, just see how optimised code looks like. Not because you have 100 functions in a C or C++ program that the generated machine code will contain that many functions. Functions can be (and are) inlined, code becomes linearised to some extent (especially on single-core microcontrollers or processors, although not necessarily limited to these) so that detecting where a function begins and where it ends is increasingly difficult. See what Matt Godbolt's Compiler Explorer produces when you write code. Turn on optimisations and see the result for yourself¹

¹ I'm talking to the audience here.

Reverse? I only saw forward engineering until I stopped watching it, because I was expecting some reverse engineering.

Where is the reverse engineering, if you give it the shematics?

What is your point? Are you even trying to demonstrate anything? Perhaps if you shared _your_ definition of "reverse engineering"² your stances would start to make some sense.

² Also mind that no one is forced to agree with yours.

9

Metrology / Fluke 732A charger

« Last post by perdrix on Today at 11:54:57 am »

I'm trying to work out why the charger in my 732a isn't switching to charge mode.   I unplugged it for a while (20 minutes), and the battery voltage fell to just below 25.3V.  I then plugged it back in and the charge light came on for a very short time and went off again.   The battery voltage is climbing slowly, but I think as a result of constant voltage charging via Q6.

Am I right in thinking that if the circuit is operating normally that R10 sets the threshold voltage for turning off constant current charging, and that R20 sets the float charge voltage.  If so how is the constant current charge turned back on? 

Why on earth is there a 15k resistor (R19) across the battery pack?

I think I'm being singularly dense today - I should be able to work this out!!

If anyone has been there, hints and tips will be most welcome.

Thanks
David

10

AI / Re: AI destroying economic system

« Last post by tom66 on Today at 11:54:13 am »

I cannot see a mechanism how this would fall down. All plausible scenarios I can come up with are so devastating that it's no different to anything else we already base our lives on (such as having electricity and clean water). But of course, me not seeing a mechanism is not a proof it doesn't exist. I would like to hear AI software development collapse scenarios, plausible and well thought ones. So far the predictions that AI generates so large and messy codebases it cannot later understand and fix them has not happened. I see no signs whatsoever it would be happening, either. It's crazily good at grepping and following code paths, running experiments and replacing small parts if something is off.

China invades Taiwan, and TSMC is destroyed by the Taiwanese military in the process? ("if we can't have it, you can't either".)

AI datacenters become overloaded with customers as they cannot maintain the ever increasing demand due to a shortage of low-nm silicon.

Price per token increases 10x and projects that were formerly viable become non-viable.

A typical Claude subscription approaches $5,000.  (The current subscription at $200 is very good value and unlikely to be sustainable long term.)