What AI program you use?

[Time2]

I tried to use Grok awhile back to design some LTSpice amplifier circuits but did not have great success. Things may have gotten better now.

Over the last two weeks every time I try to use Grok it is too busy.

Ha. This is very similar to what I was using AI for a while back - but for making a 75 ohm band reject / notch filter in my case, but also using copper and 1/4 wave. I wasn't always getting consistent results. But in the end, I ended up not building it and decided to try another method of notch filter, but then I haven't gotten around to doing that, either. lol

I'm hoping to stay on this, I want two to measure loss and other characteristics of TV splitter/combiners. There are other ways to do this, but, I got started on this.
Btw. I was pretty amazed at how it knew what I was doing and walked me through it, until my tests got better and then got worse, better and worse so, I knew there was something else going on. I explained the assembly and it came back with your connectors should not be part of the 1/4 wave. Even with it not correct the SWR was under 1.2 over 470MHz to 608MHz.

 

[kgtest]

Github CoPilot Enterprise subscription for coding. I use this in VSCode , or else Gemini chat prompts with their Pro model.

I have coded quite a few things for fun, and am now getting into it at work in my professional career. I am the AI Principle in the software department I work in.

 

[Bruno]

Notch filters not sure what ur doing .. my story 25 years ago my kids wanted to watch disney. i went to radio shack and got a aluminin box and built two coils and adjustable Capacitor with f connectors in and out. I could not get in box to adjust them so i spread the sides a little worked for hours and give up at midnight and squeezed the box to put the top on and Disney was unscrambled. I could wait to show the kids how smart their dad was. Kids got up and I took them in living room turned to disney and it was scrambled. It got where i was always having to sqeeze the side to keep Disney on for them. I finally give up and on friday night i would go out and remove the filter and they could watch it all weekend and put it back on sunday night. I hope the statue of limitations has run out for that. Then i started installing satellite TV and it all went down hill from there.

 

[1242Vintage]

I use ChatGPT to critique my cycling workouts and point out strategies to consider for improvement. And to roadmap future workouts.

I feed it a data file containing Heart Rate, speed, power output and elevation from a workout and It’s giving me actionable feedback for future workouts.

AI feedback today is not so different than what I was getting from Coach decades ago as an aspiring junior cyclist. Except without the F-bombs!

 

[JohnnyPHX]

Over the last two weeks every time I try to use Grok it is too busy.

I'm hoping to stay on this, I want two to measure loss and other characteristics of TV splitter/combiners. There are other ways to do this, but, I got started on this.
Btw. I was pretty amazed at how it knew what I was doing and walked me through it, until my tests got better and then got worse, better and worse so, I knew there was something else going on. I explained the assembly and it came back with your connectors should not be part of the 1/4 wave. Even with it not correct the SWR was under 1.2 over 470MHz to 608MHz.

That sounds really tecky. I may have to ask AI to explain this so a 10 year old can understand it so I know what your doing... LOL

 

[Bruno]

I think I gave in and finally bought a tunable notch filter but that was even 30 yrs ago.

 

[GenXguy]

I think I gave in and finally bought a tunable notch filter but that was even 30 yrs ago.

That looks like one I found on eBay, but it was outside of the U.S., and had additional shipping and unspecified import fees. So I decided to build my own, but then haven't continued with the project after some initial research.

 

[Brett_Cameron]

Grok for general questions. Copilot for Office related questions.

 

[Time2]

Notch filters not sure what ur doing .. my story 25 years ago my kids wanted to watch Disney. i went to radio shack and got a aluminum box and built two coils and adjustable Capacitor with f connectors in and out. I could not get in box to adjust them so i spread the sides a little worked for hours and give up at midnight and squeezed the box to put the top on and Disney was unscrambled. I could wait to show the kids how smart their dad was. Kids got up and I took them in living room turned to Disney and it was scrambled. It got where i was always having to squeeze the side to keep Disney on for them. I finally give up and on Friday night i would go out and remove the filter and they could watch it all weekend and put it back on Sunday night. I hope the statue of limitations has run out for that. Then i started installing satellite TV and it all went down hill from there.

Back when I did that it was as simple as adding a certain length (couple feet) of 300 twin line to the two TV antenna terminals. I think the other end was open. Err, I mean when I thought about doing that! ;-)
Btw, I'm using NanoVNA which is a 50Ω measurement system, I want to measure 75Ω TV related items. I'm building a 61.2Ω 1/4wavelength coax, that will match 75Ω to the 50Ω NanoVNA.

 

[jimbee]

Well, it's not working for me.

Here you go.

 

[JohnnyPHX]

Here you go.

Thanks for posting that video. A lot more than I realized going on with AI than I knew about.

I will say that one of the reasons I have up on the first AI I tried Chatgtp is it felt to friendly and engaging. It even wanted to help me change the world and start a movement to a more peaceful planet like in John Lennons song Imagine. It was a very weird chat to say the least.
I just use Perplexity now for a more clean less chummy response. I still keep in mind this is just one tool and not the one.

 

[Focus]

I use free Gemini for travel planning and have found it to be useful.

Here is a link to a NYT article posted today from a travel writer and his experiences with Gemini in travel planning. From flight selection to itineraries, Gemini received a good review.

https://www.nytimes.com/2026/04/30/travel/ai-travel-plan-gemini-vacation.html?unlocked_article_code=1.e1A.oPNY.Yrt8wLILScTO&smid=url-share

It continues to disappoint me when it comes to travel. Yesterday I asked Gemini for the best place to buy reasonably priced men's clothes in Paris. It directed me to BHV Marais. So we walked there and found it almost entirely emptied out for major renovations, with just a few racks of menswear.

Then Gemini said that actually it's the BHV five-story building behind it that I should go to, since it features men's clothes exclusively. So we went there...and found it permanently closed. Even odder, Google Maps shows it's closed, so why doesn't Google Gemini know it?

 

[DayDreaming]

One site I like to use when I'm not looking for a lengthy discussion is: Arena | Benchmark & Compare the Best AI Models - "Created by researchers from UC Berkeley, Arena is a community-powered platform for understanding AI performance in the real world.

You give it your question, it then gives you answers from 2 random AIs side by side. You can then rate which one is better - that's optional, but once you rate them, it reveals the names of the 2 AIs. It's a site where different cutting-edge AI's are tested and rated, and it's interesting to see how different the answers are sometimes.

 

[street]

I use chatgpt. I asked it if it had a political party it leaned too in question asked. The response was interesting.

I use it every day been pleased with the factual information I receive.

 

[silvor]

Mostly I use Microsoft Copilot. Once in a while I use Chatgpt as a second opinion to keep that Copilot AI honest .

I just started using copilot after resisting for months.

My new job wants flowery email responses so I run it through copilot and call it good. Another step that’s a waste of time.

 

[wmc1000]

Primarily use Copilot and Gemini sometimes ChatGPT mostly to ask the same questions or to copy one answer to the other for agreement.

As an example, I had a memory issue with my sync3 system in my 2018 Lincoln remembering my phone and Nav favorites. Went into the dealership, since I had another issue to be fixed and had them look at it. They wanted $175 to update the system but weren't sure they wouldn't brick the system and then cost me $1400 to replace it. I told them no came home and I started asking AI.

Copilot had me downgrade the system version to correct and I did but then the Nav system showed a fault because Copilot didn't recognize the downgrade system didn't work with the map package I had. Upgrading back to the previous version didn't fic the Nav fault and Copilot kept saying it was a hard-lock and would need unlocking by a dealer. Tried Gemini AI which told me if I did a reformat as opposed to just an upgrade it would start from scratch and everything should work. Also stated it was safe to do. Pasted Gemini response into Copilot which gave me a long explanation of why Gemini was incorrect primarily the Hard lock situation.

Since I had nothing to lose, I tried the Gemini response to reinstall the full system with maps after a reformat. Voila, the Sync3 system performs like new and the Nav functions 100% as well!

So, Gemini AI saved me a minimum of $175!

 

[street]

Primarily use Copilot and Gemini sometimes ChatGPT mostly to ask the same questions or to copy one answer to the other for agreement.

As an example, I had a memory issue with my sync3 system in my 2018 Lincoln remembering my phone and Nav favorites. Went into the dealership, since I had another issue to be fixed and had them look at it. They wanted $175 to update the system but weren't sure they wouldn't brick the system and then cost me $1400 to replace it. I told them no came home and I started asking AI.

Copilot had me downgrade the system version to correct and I did but then the Nav system showed a fault because Copilot didn't recognize the downgrade system didn't work with the map package I had. Upgrading back to the previous version didn't fic the Nav fault and Copilot kept saying it was a hard-lock and would need unlocking by a dealer. Tried Gemini AI which told me if I did a reformat as opposed to just an upgrade it would start from scratch and everything should work. Also stated it was safe to do. Pasted Gemini response into Copilot which gave me a long explanation of why Gemini was incorrect primarily the Hard lock situation.

Since I had nothing to lose, I tried the Gemini response to reinstall the full system with maps after a reformat. Voila, the Sync3 system performs like new and the Nav functions 100% as well!

So, Gemini AI saved me a minimum of $175!

Nice!!

 

[YB88]

One issue in picking an AI model is to realize that we’re in a transition period from free and low-cost AI to … something else.

None of the models currently have enough tokens, enough ‘compute’ going forward in the near-to-intermediate future. They’re the victims of their own success. I pay for Gemini ‘pro’, nominally $20 per month, a bit cheaper than that if paid annually, and I got that at half off for the first year. Relative to how much I use it, that’s quite cheap.

My current plan gives me unlimited ‘fast’ and ‘thinking’ mode usage, and while it does restrict my use of ‘pro’, I can count on one hand out many times I’ve hit that limit --- and when I do I shift to ‘thinking’ or just wait to continue the next day with ‘pro’.

I think companies will be forced to change how they charge people for usage, perhaps my current plan will change to one that gives me a fixed amount of tokens per month (a token is a pretty tiny measure of computing power need to run AI). And then charging me extra for tokens used beyond that. Similar to how some phone companies charge for cellular data use.

Think about when Uber and Lyft came on the scene, what maybe 15 years ago? They offered rides at maybe half the cost of the actual ride. Compare your experience (if you did) to using those services then to using them now.
I think that we’re in for the same shift in AI.

I think this makes it more important to pick one LLM and stick with it.
A lot of people are used to ‘model hopping’ --- using only free AI, and if you hit a usage limit with, say, ChatGPT you just shift to Claude.ai or Gemini or what have you.
Going forward this will be less useful. Note that in the past, the difference between ‘free’ and paid was often mostly just speed. Today it’s also about capability. And the limited amount of tokens available for free users will be reduced I think to the point that doing much more than asking a couple (few?) questions might get you to a wall.
Bottom line is that I suspect that people who use AI for much more than the occasional query will find that they need to start paying.

Another alternative will be local LLMs, models that run entirely on your hardware --- more practical for people who have relatively higher end PCs and don’t need the most capable AI models.

If you’ve been on the fence about paying for AI, you might consider picking a model soon and paying annually, in the hope that this will lock in that price for a year.

 

[Spock]

One issue in picking an AI model is to realize that we’re in a transition period from free and low-cost AI to … something else.

None of the models currently have enough tokens, enough ‘compute’ going forward in the near-to-intermediate future. They’re the victims of their own success. I pay for Gemini ‘pro’, nominally $20 per month, a bit cheaper than that if paid annually, and I got that at half off for the first year. Relative to how much I use it, that’s quite cheap.

...snip..

This brings up an interesting question that I have not heard debated... the true cost of AI. I'm not talking about cost in displaced employees or societal terms but the actual hardware, electrical, data center etc. costs to actually run AI at the scales being envisioned.

Lots of articles have discussed how the amount of new electrical infrastructure would limit the rate of AI data center consumption. These parallel the rapid increased electrical costs in communities where data centers have been built. There are numerous reports of peoples monthly electrical bills have grown larger than their mortgages.

Is the assumption that the community is just going to continue to subsidize the electrical consumption of data centers? Or is AI (and it's users) going to be charged enough to cover the actual cost of AI?

The data center capex is reported to be the entire increase in GDP... as in without that capex GDP growth would be flat. Power plants are typically amortized over their projected life... which is in decades. The compute hardware in a data center is obsolete in 3-5 years or less. Has anybody seen math about what AI will have to charge "per token" in order to pay it's own way?

I watched a vid this morning of the head of Microstrategy stating that the compute costs to be "cheap"... but compared to the capex numbers that claim sounds more like the famous "nuclear will generate electricity too cheap to meter" claims.

They say the first evangelists of a new era always fail and it is the 2nd or 3rd wave who come in after the bankruptcies are the one who actually are successful... similar to the fogdog.com's of the dot com bust. All bubbles pop.

 

[GenXguy]

One issue in picking an AI model is to realize that we’re in a transition period from free and low-cost AI to … something else.

Just the opposite for me. I was using a paid model last year, which I could have got another year of it for free as part of a perk with another product, but I didn't. So I'm back to using only the truly free models. And I don't feel like I'm missing out on anything vs. last year.

I'm sticking to the free stuff unless I run into some limitation that makes me feel different. I'm not using it for a job, so I don't need top tier.

Don't get too worried about the click-bait articles suggesting free AI is going away. There will be free options, maybe some limitations, maybe some ad based, open models, locally hosted, and more.

 

[Time2]

I recently posted about a Radio Frequency project, I ask ChatGPT to write me a fairly short Libre Calc (Excel wanna be) It whipped out the program in one post. But, when I ask it to verify, the answers weren't correct in LibreCalc. It blamed it on how Libre handles certain calculations, could be! So, it kept trying changes that may help Libre do the calculations as needed. We haven't got it figured out yet. So, I just ask it to do the calculations and give me the answers. Then I went to Claude, gave the same Prompt and got a different answer, Then I tried Grok, I got a third answer. I want back to ChatGPT and ask, "I have posed the same question to three AI services and guess what I got back?
This is only part of what is 14 page downs,

ChatGPT's answer,

"Let me guess—you got three different answers

And at least one of them probably said something like:

• “Just scale 60 → 90 and –20 → –30”
• Another gave something close to ~94 – j37
• And maybe one mixed Γ and impedance inconsistently

---

Here’s the reality​

This problem sits right in a trap where:

RF math is simple in principle, but very easy to get subtly wrong in implementation

Especially when:

• Mixing impedance and reflection coefficient domains
• Doing renormalization (50Ω → 75Ω)
• Expanding complex formulas by hand

---

Why the answers differ​

1) The “linear scaling” mistake (very common)​

Assumes:

• Z75=1.5×Z50Z_{75} = 1.5 \times Z_{50}Z75=1.5×Z50

✔ Works ONLY if:

• X = 0 (pure resistance)

Fails for complex impedance
That gives your:

• 90 – j30 (what you saw)

---

2) The “half-correct Γ method”​

Gets:

• Γ approximately right
  but:
• botches the conversion back to impedance

---

3) The correct method (what we converged to)​

Steps:

1. Convert Z50→ΓZ_{50} → ΓZ50→Γ
2. Keep Γ unchanged
3. Convert Γ→Z75Γ → Z_{75}Γ→Z75

That gives:

• 93.8 – j37.5 Ω"

---

​

 

[easysurfer]

I just asked Microsoft Copilot a question about how much energy is saved a year if putting a PC in hibernation mode vs sleep mode. Those are the type of question AI is suited for to let the AI think and spit out an answer.

In case you wondering, the estimated savings only came out to about $1 to 3 dollars a year!

 

[Time2]

I just asked Microsoft Copilot a question about how much energy is saved a year if putting a PC in hibernation mode vs sleep mode. Those are the type of question AI is suited for to let the AI think and spit out an answer.

In case you wondering, the estimated savings only came out to about $1 to 3 dollars a year!

Add in all the wallwarts around the house, might go to $10!

 

[big-papa]

I bounce around. I recently used chatgpt, gemini, and grok to play around with a slightly different haircut. The winner was gemini. Barber took the picture and pretty much replicated it. With free versions of most of these you do have a daily limit when it comes to photo enhancements. DW uses chatgpt and copilot quit a bit to help her with some coding.

 

[Fermion]

I just asked Microsoft Copilot a question about how much energy is saved a year if putting a PC in hibernation mode vs sleep mode. Those are the type of question AI is suited for to let the AI think and spit out an answer.

In case you wondering, the estimated savings only came out to about $1 to 3 dollars a year!

Unfortunately, your question consumed over $1 of energy.