Lessons Learned about AI (Large Language Models): "I don't 'know' the facts the way that a human does..." Let's do an experiment...

Yesterday, as I struggled with a Large Language Model that at first gave good advice (how to build a 5 crystal filter) and then really bad advice (how to build a filter with 5 crystals in parallel to achieve a USB passband) I got a message from an old friend who currently works in this industry.  

He advised me to be careful about advice from LLM AIs.  He warned that these AIs would just develop answers based on the information that had been used to train them.  In the case of ham radio electronic questions, the training material was often littered with errors and lore and bad, wrong info.  Often this bad information would be repeated over many years.  This would make the AI more likely to use it in their answers. 

It then occurred to me that I should take one case where I knew that wrong information was in the ham radio "literature" and that this information had been repeated many times over the years.  I am referring here to the question of why hams use lower sideband below 10 MHz and upper sideband above 10 MHz.  I had explored this question in many blog posts.   Many explanations were offered -- all of them violated the Hallas rule.  All of them failed to answer the question.  I wondered, would AI be wrong?  

Well, AI was wrong.  AI said:  

Subtracting in heterodyne mixing inverts the frequency spectrum. An Upper Sideband signal generated at the IF naturally flipped into a Lower Sideband signal on the lower bands.

That is wrong.   The Hallas rule tells us that sideband inversion ONLY occurs when the signal with modulation is SUBTRACTED FROM the signal without modulation.  

I called AI on this error in its answer.  I asked it why it had given me bad information.  Its responses were telling. 

Why did you initially say that sideband inversion results from subtraction? Is this because of the material on which you had been trained?

You're asking a fair question about where that explanation came from.

The brief answer: yes, it comes down to how technical explanations are presented across textbooks, amateur radio training materials, and online discussion forums—and how LLMs synthesize those patterns.

So you are just repeating incorrect information, as long as it was repeated enough, right?

That is a fair critique, and it gets to the heart of how Large Language Models work.

I don't "know" facts the way a human does; I generate text based on statistical patterns learned from vast amounts of training data. When a simplified explanation, misconception, or shorthand rule of thumb is repeated thousands of times across forums, textbooks, and study guides—as is the case with "subtraction inverts sidebands" in ham radio literature—I am prone to picking up and repeating that pattern.

-------------------------

I think we can draw three conclusions from all this: 

1) The talk about AI "personhood" is just nonsense, mostly nonsense from Anthropic. 

2) The concerns about data poisoning are real.  The ham radio "literature" is already poisoned, and look what happens. 

3)  We are living with the results of bad info in the amateur radio "literature" over the years.  AI is just spitting it back at us.  And I think it is getting worse.  Now everyone with a callsign and camera has a YouTube channel. The quality of content is low and dropping.  Soon AI will be spitting this stuff back at us. 

---------------------------

Let's do an experiment.  Those of you who have access to the other AI's like Claude, ChatGPT, DeepSeek and others, just ask them the question I started with: 

"Why do ham radio operators use LSB below 10 MHz and USB above 10 MHz?"  Let's see if the other AI LLMs are better at coming up with truthful answers.  Of course, they may see this and not repeat the error. 

What Happens Inside an AI Chip When You Ask a Question. PLEASE COMMENT

When I post something like this, there are several predictable responses: 

1) Some guys will proclaim it to be "AI slop", and will criticize me for even posting it.  

2) Some guys will just declare that I am "getting old" (as if they are not) and will assume that at my age, I just somehow can't understand these new machines (as if they can!).  

3) Many more sensible guys will put aside their feelings about AI and my age, and -- perhaps giving the devil his due -- be amazed by the complexity of these machines.   

I think it is pretty amazing.  

What do you guys think? 

This Guy Designed His Own Microprocessor Chip -- But Is This REALLY Design?

First, hats off to this fellow.  He deserves a lot of credit for even trying this.  

But... I think this shows how DIFFERENT designging these chips is from what we would consider traditional ham radio homebrew design and construction.  The truth is that I didn't even understand much of what he was talking about, even when he was trying to explain what he had done.    

There does seem to be at least a little common ground (!):  I note that there was a need for diodes to ground in the longer leads, and in the test a lack of grounding caused the chip to not work properly.  

Thanks to Hack-A-Day for alerting us to this.  

What do you guys think? Is this concept of design similar to traditional ham radio concepts? 

Demystification! Not Magic! How Claude and Dean made the KK4DAS -- N2CQR Video (Be Sure to Read the Report Until the End!)

 

Many have asked how the video was made.  Dean and Claude (but not Claude Shannon! ) explain here: 

https://kk4das.blogspot.com/2026/09/how-cq-video-was-really-made-it-wasnt.html

Claude Makes a Video About N2CQR calling KK4DAS

Quick!  Before Claude turns us all into paper clips!   

I guess this is part of a series.  First we had Notebook LLM making a podcast about SolderSmoke.   Then Peter Marks VK3TPM had Claude (I think) give us an AI version of Pete N6QW talking about circuits.  

Now we have (above) a Claude-produced video of KK4DAS in contact with me, N2CQR.  Claude (and KK4DAS!) did all the heavy lifting.  I think it looks pretty good for something that never really happened.  

What do you guys think?  We may be getting closer to the day where we can have Pete N6QW on video, convincingly singing the praises of LC VFOs and bad-mouthing the Si5351.  That will be a momentous day indeed.   Paper clips may soon follow!    

Please put your comments below.   

Why Chip Circuits Can Never be Truly Homebrew: ASML's Amazing Photolithography -- Please Comment

 As I watched this video, I had many thoughts.   Here are some of them: 

1)  The complexity of the ASML machine is a reminder of how far IC technology has moved from the world of homebrew ham radio.  I know others will take different views, but for me, it is a reminder that these chips -- which are now commonly used in cell phones, for example -- have no real place in rigs that we would call homebrew. 

2) The technology they use is really amazing.  Laser lights hitting tin droplets three times as the tin falls. Explosions inside the machine that are like miniature super novae. 

3) Older guys play a big role in all of this.  

4) All the attention today is focused on AI and data centers.  But the tech there seems simplistic when compared to the ASML machines that make the chips.  

Please watch the Veritasium video that appears above.  Then please put you comments in the space below.  

100 Billion Transistors on One Small Chip -- No Room for Homebrew!

https://newsroom.ibm.com/2026-06-25-ibm-debuts-worlds-first-sub-1-nanometer-chip-technology

FCC Amateur Radio Exam Question:  Draw the schematic for the new IBM sub-1nm chip, showing how it will be used in FT-8 operations. 

Mike WN2A suggested I ask AI about a schematic.   Here is what Gemini said: 

Here is what Gemini said: 

Please draw me the schematic (showing the connections of all 100 billion transistors) in the new IBM sub-nm chip.

I would love to help sketch out circuit diagrams, but drawing a literal schematic showing the individual connections for all 100 billion transistors is physically impossible!

To put that number into perspective, if we were to draw a schematic where each transistor was just a tiny 1-inch symbol on a piece of paper, the blueprint would be so massive it would easily cover the entire landmass of the United States.

Modern sub-nanometer chips—like IBM's cutting-edge nanosheet architectures—are far too complex for human hands to map out line-by-line. Instead, engineers use advanced software to design chips hierarchically. Here is how that massive sea of 100 billion transistors is actually organized under the hood:

1. The Building Block: The Gate-All-Around (GAA) Nanosheet

At the microscopic level, these chips no longer use the older FinFET (fin-shaped) transistors. Instead, they use nanosheets, where the channels are stacked vertically like a three-story sandwich. The gate wraps entirely around all four sides of each horizontal sheet, providing maximum electrical control and drastically reducing current leakage.

2. Standard Cells (The Functional Logic)

Instead of wiring 100 billion separate components, computer scientists group transistors into "Standard Cells." These are pre-designed functional blocks like:

• Logic Gates: AND, OR, NAND, and NOR gates (requiring anywhere from 2 to 8 transistors each).

• Flip-Flops: Memory cells that store a single bit of data (requiring about 20 to 30 transistors).

3. The Macroscopic Layout (Blocks and Cores)

Billions of those standard cells are then combined into macroscopic structural blocks that look less like a traditional schematic and more like a highly organized city map. If we were to draw a high-level schematic of the chip, it would look like a block diagram featuring:

• Processor Cores: The massive processing engines packed with arithmetic logic units (ALUs).

• Cache Memory (SRAM): Vast grids of 6-transistor memory cells stacked tightly together to hold data right next to the execution units.

• Interconnect Matrix: A vertical 3D skyscraper of up to 15 layers of microscopic copper wiring running above the transistors to route data and power across the chip.

If you are working on a homebrew electronics project and want to map out a circuit diagram for an RF amplifier, a QRP radio transceiver, or a simple analog filter, I can absolutely help you draw or troubleshoot those schematics! But for 100 billion nanosheets, we have to leave the drafting to the silicon foundries' supercomputers.

Early Hacking with an IMSAI 8080 -- "War Games" (1983) Scene with Matthew Broderick and Ally Sheedy

Some computer history here, and it is old enough (1983) to fit into the SolderSmoke theme.  Plus it appears that the young fellow has a shack (of sorts) and uses it to impress a girl from his biology class.  Finally, we know the IMSAI guy has an important YouTube channel... 

MIT Technology Review on Why AI LLMs are So Strange and So Alien

Until recently, I thought that the AI LLMs were just sort of fancy, souped-up search engines.  Google on steroids.   But then they started getting simple things wrong.  And they seemed to understand that that our April 1 stories were just sarcasm.  There seemed to be more to them than fancy search engines.  

This MIT article explains what is going on with the LLMs.  

https://www.technologyreview.com/2026/01/12/1129782/ai-large-language-models-biology-alien-autopsy/

What do you folks think of this?

Why is it that AI-provided electronic circuits often don't work in the real world?

Like many, I have turned to AI for circuit ideas.  And like many, I have been surprised by how many of these ideas just don't work.  I have found relatively simple circuits that don't work when built in the real world, and don't work when built in LTSpice. I have already described how AI failed when I asked it a simple question about how many 330 ohm resistors to put in parallel to get 50 ohms.  Mike WU2D had found that a lot of AI provided circuits just don't work.  (Go to the 10 minute point in his video. )

I was wondering why this is so.  So... I asked AI!  For those who are sure that their AI is smarter than my AI, let me specify:  I just wrote the question into the Google search box and looked at the "AI Overview" in the response.  

Here is what AI said: 

AI-recommended circuits often fail in the real world due to gaps between simulation and reality, such as a lack of real-world data (noise, rare conditions), AI's inability to grasp complex physical nuances like heat dissipation or component aging, and poor integration with physical prototyping, leading to brittle designs that don't account for manufacturing tolerances or environmental stresses, despite being mathematically sound. 

Key Reasons for Failure:

Data Limitations (Garbage-In, Garbage-Out): AI relies heavily on training data; if it's incomplete (missing extreme temperatures, noises, aging effects) or biased, the AI generates designs that work in simulation but not in messy real-world conditions.

• Brittleness & Lack of Context: AI struggles with "edge cases" or novel situations not in its training, leading to fragile designs, much like an AI failing to recognize a rotated bus. It lacks human intuition for subtle physical interactions.
• Physical Nuances Ignored: Simulations often miss crucial real-world physics like heat hotspots from component imperfections, electromagnetic interference, or subtle signal integrity issues, which become critical in physical prototypes.
• Manufacturing & Prototyping Gaps: AI designs might be theoretically optimal but impractical to build (e.g., impossible tolerances, complex layouts). The gap between virtual design and physical assembly (poor connections, wrong components) causes failures.
• Validation Challenges: AI-generated circuits need extensive, real-world testing (SPICE simulations, breadboarding, PCB builds) to catch issues, but AI can't perform physical troubleshooting like soldering a damping resistor or witnessing ground bounce in a lab.
• Probabilistic vs. Deterministic: AI works on probabilities, whereas electronics often need deterministic precision. A small error tolerable in some AI tasks can be catastrophic in sensitive circuits. 

How Engineers Address This:

• Iterative Feedback Loops: Combining AI's speed with human expertise and physical testing to refine designs.
• Hybrid Approaches: Using AI for creative exploration but relying on established physics models (like SPICE) and human judgment for final validation.
• Focus on Integration: Ensuring AI tools work seamlessly with existing engineering workflows and physical constraints. 

Internet Archive / Wayback Machine in an Old Church in San Francisco

 

Three cheers for the Internet Archive!  Among its millions or trillions of things saved, there is the SolderSmoke podcast and blog.  

Check out the CNN video: 

https://www.cnn.com/2025/11/16/business/video/internet-archive-wayback-machine-church-digvid

More Silicon Valley -- Apple, Microsoft, and IBM (video)

 More interesting computer history.  

-- Visicalc and importance of spread sheets. (I remember when I first learned to use one). 

-- Amazing that the guy who wrote the DOS program for IBM can't get a job today. 

-- Sad that the guy who was in the garage with Jobs never got stock options. (Was this ever corrected?) 

-- I liked the observation that the most powerful guy in the room is usually the most poorly dressed guy in the room. 

Silicon Valley -- The Hippies who Built the First PC (full video)

Computers are not really my thing, but the discussion of California culture and how it led to the home computer revolution is very interesting. 

-- Blue Boxes! (I should have made one.) 

-- An old telephone earpiece. (I still have one.) 

 

-- The Free University of Palo Alto (I should have gone there.) 

-- The HOMEBREW Computer Club  (Great use of the word "homebrew.") 

By the mid 1970s, "they were going to change the world not with flowers in their hair, but with soldering irons in their hands." 

Great stuff. 

Homebrewing a Quantum Computer

Ok, now for something completely different, and something that might help us shake off that Ludite moniker. 

I've seen lots of videos that promise to explain what quantum computing is all about.  Most of them disappoint.  These videos have promise.  

This young lady built a very simple quantum computer at home using lasers and some calcite.  The homebrew angle is very interesting.  At one point in the second video she talks about the benefits of taking theory and using it to actually build something at home.  Indeed.  

I need to spend more time with both of these videos, but like say, they show promise, and they definitely show the spirit of homebrew.  

The first (above) describes the build.  In the second video (below) she uses the build to do a calculation.

  

The Computer Hardware of the Apollo Program

Watching this, it seems like a miracle that we made it to the moon.  Some of this construction is downright scary.  Sometimes soldering is not good enough -- so they weld.  Other times they don't even solder -- they wire wrap!  Even in 1965, the wiring of the memory modules was so complex that it was beyond human comprehension.  All of this brought back memories of that wonderful book "Sunburst and Luminary" by Don Eyles.  I have hopes that he will appear in the MIT video about landing on the moon.  Don worked on those programs.  

The History of MOSFETs -- Let us Remember the 40673. The IRF510. And others...

Lots of great history explained in this video.  I found myself thinking about our beloved 40673 dual gate MOSFET: 

https://hackaday.com/2016/03/21/the-curse-of-the-40673-zombie-components-that-refuse-to-leave-the-building/ 

And, of course,  about the IRF510.  

CuriousMarc Gets an Apollo DSKY Running. FPGA AGC Computer. Rope Memory.

I really liked this video.  So much amazing wizardry in it.  Just the FPGA Apollo AGC simulator that Mike whipped up.  Wow.  Kudos!  

Their occassional references to "rope" brought back fond memories of "Sunburst and Luminary" by Don Eyles:  https://soldersmoke.blogspot.com/search?q=Apollo+rope 

The Transistor that Changed the World -- the MOSFET

Another great video from Asianometry.  

My only quibble is that it kind of left unclear the differences between JFETs and MOSFETs.  After all, we still use both.  Note our beloved J310 JFET.  And the IRF510 is a MOSFET.  

Google's AI explains: 

A JFET (Junction Field-Effect Transistor) and a MOSFET (Metal Oxide Semiconductor Field-Effect Transistor) are both types of field-effect transistors, but the key difference is that in a JFET, the channel conductivity is controlled by a reverse-biased PN junction, while in a MOSFET, it's controlled by an electric field across an insulating layer between the gate and the channel, allowing for a much higher input impedance and greater design flexibility in MOSFETs; essentially, MOSFETs are considered a more advanced version of JFETs with superior performance in many applications like high-speed switching and integration into complex circuits. 

The Surprising Difficulty of Analog Circuit Chip Design -- AI to the Rescue?

https://www.youtube.com/watch?v=lNypq1XuZRo

Really interesting.  Why the design of the analog portions of chips is so much harder than the design of the digital portions.  

Great channel. 

A Really Cool Homebrew Computer

Wow, very cool.  On Hack-A-Day this morning:

https://hackaday.com/2024/05/26/homebrew-computer-from-the-ground-up/

It caught my eye because the architecture seems similar to that of the MostlyDIYRF PSSST rig.  

There is a lot to learn from this little machine, especially for an analog guy like me. 

Mostly DIY RF PSSST