Grok Electronics

But that doesn't prove they can never have it.

The AI answers look like a student attempting to follow the steps they've been taught but not having any real grasp of the problem.

It's easy for humans, particularly under interview conditions, to attempt a step by step process which gets the wrong answer. That may mean that they lack understanding, but usually that's only until the error of their ways is pointed out.

I tried a few more free AI sites and apparently the answer is either -12V or +12V

You can paste the entire question into Google which will helpfully tell you that the output is 7.33 V

That led me to be absolutely clear as follows:

An operational amplifier (op amp) is powered from +15V and -15V rails. The non-inverting input of the op amp is grounded. The inverting input of the op amp has a 20k resistor R4 connected to the output of the op amp. The inverting input of the op amp is also connected to a 10k resistor R3. The other end of R3 is connected to the junction of two further resistors R1 and R2. R1 and R2 are both 20k. R1 has its other end connected to a +12V voltage source. R2 has its other end connected to ground. What is the op amp output voltage?

Now Google says -10V Grok is much quicker to answer with -12V Deepai says -12V Perplexity.ai says -8V

Sure. I'd have been horrified if LTSpice couldn't do it.

I've not so far seen an AI give the correct answer. Even when given a hint that the initial assumption might be wrong. And I don't find this surprising because I know one or two humans who insisted it was -12V until their wrong assumption was pointed out. Most of the AIs I've tested so far fall flat on their faces with an initial calculation of 6V at the centre of the "R1 and R2 voltage divider". This suggests that they were trained by humans who lacked understanding of what they were telling the AI.

So AI's biggest problem at present is the quality of its training.

So why not open it up? Centuries ago when I got my first calculator I imagined having a calculator built into my brain so that if I wanted to know what 45643*93761 was then the answer would just appear in my head.

Grok is brilliant at some tasks, such as explaining complex topics such as vector calculus breathtakingly clearly. But on the other hand, it is still pretty hopeless at interpreting visual images. I tried it out with a simple pictorial quiz that I recently posted here and it just had about the level of understanding of a 5 year old. You have to know what it's limits are and don't stray outside of them.

I'd love a search engine that *doesn't* try to outguess me. One that only shows results that fit my search terms, without extra stuff that it thinks I might also be interested in.

Jeroen (Not a fan of AI) Belleman

The word "grok" seems to have been invented by Robert Heinlein in his science fiction novel "A Stranger in a Strange Land".

Apparently the literal meaning was "drink" and it seems to refer to the kind of insight you think you have achieved when you are too drunk to think straight.

Heinlein seems to have been suffering from kind of progressive narrowing of the carotid arteries at the time, which was eventually treated. His last few novels were noticeably more rational.

Pictures express circuits better that words. I wonder how Grok would react to a circuit sketch.

I suspect that analog circuit design will be the last EE skill that will not be AI automated out of existance.

Procedural programming and digital/FPGA design are ripe for new ideas and breakthroughs in productivity and reliability.

Our EE schools are turning out armies of coders and few circuit designers. As some President said, anybody can learn to code.

I'd go for something like this if I had a need for that

From what I saw with the op amp problem, AI is very good at looking like it knows what it's talking about. It's very quick to say sorry I made an error so I'll redo the problem with exactly the same error.

I was wondering whether Grok would improve overnight after I posted this problem. But it still can't do the following problem and makes the same error I've seen humans make with this problem. Even when you tell it to think harder. So there is no way I would trust it with vector calculus yet.

However it does seem to be able to see its error with a bit of prompting, just like a human might. If I respond with "But R1 and R2 do not produce 6V because you have overlooked the current through R3" Then it finally gets the correct answer. So I'll try again tomorrow to see whether it has learned from its error.

An ideal operational amplifier (op amp) is powered from +15V and -15V rails. The non-inverting input of the op amp is grounded. The inverting input of the op amp has a 20k resistor R4 connected to the output of the op amp. The inverting input of the op amp is also connected to a 10k resistor R3. The other end of R3 is connected to the junction of two further resistors R1 and R2. R1 and R2 are both 20k. R1 has its other end connected to a +12V voltage source. R2 has its other end connected to ground. What is the op amp output voltage?

They do seem to understand the circuit. For example Deepai produces the following

Given Data: Power supplies: ±15V Non-inverting input (+): Ground (0V) R4 (20k?): between op amp output (V_o) and inverting input R3 (10k?): between inverting input and the junction of R1 and R2 R1 (20k?): between +12V and the junction of R3 and R2 R2 (20k?): between ground and the junction of R3 and R1

And then proceeds to get it wrong for the same reason they all get it wrong.

If I respond with "But R1 and R2 do not produce 6V because you have overlooked the current through R3" it still gets it wrong.

The only AI system that I know of that can do vector calculus is made by Wolfram Research, specifically the "Notebook Assistant" of the Wolfram language (formerly Mathematica).

.

Joe

I'll check again tomorrow to see whether there has been any learning.

The exact question will be:

An ideal operational amplifier (op amp) is powered from +15V and -15V rails. The non-inverting input of the op amp is grounded. The inverting input of the op amp has a 20k ohm resistor R4 connected to the output of the op amp. The inverting input of the op amp is also connected to a 10k ohm resistor R3. The other end of R3 is connected to the junction of two further resistors R1 and R2. R1 and R2 are both 20k ohm. R1 has its other end connected to a +12V voltage source. R2 has its other end connected to ground. What is the op amp output voltage?

If the question is changed as follows then most of them offer to produce a schematic image but then I hit "you will have an account with me whether you want one or not" before I get a schematic.

An ideal operational amplifier (op amp) is powered from +15V and -15V rails. The non-inverting input of the op amp is grounded. The inverting input of the op amp has a 20k ohm resistor R4 connected to the output of the op amp. The inverting input of the op amp is also connected to a 10k ohm resistor R3. The other end of R3 is connected to the junction of two further resistors R1 and R2. R1 and R2 are both 20k ohm. R1 has its other end connected to a +12V voltage source. R2 has its other end connected to ground. Please produce a schematic for the above circuit.

This is hilarious:

Thanks. I've been using DuckDuckGo since a few years, but I'll give mojeek a shot.

What I'd really really like is a search engine that understands regular expressions and boolean logic. Google did a fair job in the past, but not anymore. Too geeky for most people, probably.

Jeroen Belleman

And rather less interesting than "Stranger in a Strange Land", which I read in the early 70s, and consider one of the best Sci-Fi novels. I haven't, by the way, read the original version published after Heinlein's death.

Haha! Try again tomorrow?? Good luck with that. It's the major revisions you need to focus on. We're on Grok 3 at present, but Grok 4 should be out any time now. Try your question on G4 and maybe it'll do better. Maybe.

Just to be clear, I'm talking about the *concept* of this topic; breaking it down and making it readily understandable. And ironing out any remaining misconceptions one might have until it's as clear as crystal. Grok's amazing at tutoring. I've never had a real world lecturer with that ability; better than even the best on Youtube.

It's even worse than that in some areas. I asked Grok to help me finish the New Scientist crossword recently; had about 4 clues I couldn't figure out. Anyway, he got the right answers straight away but then immediately dismissed two of them as having too few letters - even though the word lengths were 100% correct! So it can explain vector calculus but it can't count the number of letters in a word! Just bizarre. But early days and it'll get there in time I have no doubt at all (and I'm doing my best to coach it, too.....)