Another AI thread

[Mr Winterville]

I read this today (link below) and most of it wasn't news to me, except the part about AI being used to write itself now. I work in software engineering and have been an avid user of AI for a while--mainly using the models from Anthropic/Claude lately. At first it was good for writing short functions, but now if you give it the right prompting, it can build out full applications almost instantly. You can also feed it a ton of code, multiple documents, etc and it can find the bug almost instantly. So at this point it "helps" me in my job, because I'm the one asking it the questions. At some point though, it can and will be making the higher-level architectural decisions.

Something Big Is Happening

A personal note for non-tech friends and family on what AI is starting to change.

shumer.dev

 

[Heel1988]

Scary stuff but I thank you for sharing. I am trying to get my kids to understand what is coming (or is already here).

 

[Mr Winterville]

From a chat I had with Claude Sonnet 4.6 where I had mentioned performance sometimes having an inverse relationship with effort:

"To your question about whether I can relate — not experientially, since I don't have motor programs or anxiety. But I do something analogous in a loose sense: my best outputs tend to emerge from letting the process run rather than over-constraining it with explicit rules mid-generation. Whether that's meaningfully similar or just a superficial analogy, I'm genuinely not sure."

And this, in response to a question about how it "reasons" through complex questions:

"The short answer is that I'm built on a transformer architecture trained with a combination of techniques that enable something that looks a lot like reasoning — though the exact mechanisms are still an active area of research, even at Anthropic."

That last line is a bit concerning. It's not the first time I've heard/read that either. I seem to remember an interview with one of the first pioneers of AI saying that they (researchers) were not sure how the models work at anything below the highest level. And this was a couple of years ago, so the models they have today are definitely even less understood by the "researchers".

 

[PINEHEEL]

I have been a ChatGPT subscriber for about a year now and to be honest I barely scratch the surface with it. I use it as a glorified google search engine. That said, I've become increasingly frustrated with it lately as it has straight up lied to me about a couple of things, and though I prompt it to not answer confidently unless it is and to use brevity more often, it ultimately results back to its default eventually.

I haven't spent any time time playing with Anthropic, Gemini, Grok, etc. and am wondering if I should consider switching to another platform? I don't like the idea of learning a new system or starting from scratch on familiarizing a new system with my history, but I also don't want to be using an inferior system just because I'm too lazy to change.

Thoughts or suggestions from anyone who has used multiple models?

 

[Mr Winterville]

I have been a ChatGPT subscriber for about a year now and to be honest I barely scratch the surface with it. I use it as a glorified google search engine. That said, I've become increasingly frustrated with it lately as it has straight up lied to me about a couple of things, and though I prompt it to not answer confidently unless it is and to use brevity more often, it ultimately results back to its default eventually.

I haven't spent any time time playing with Anthropic, Gemini, Grok, etc. and am wondering if I should consider switching to another platform? I don't like the idea of learning a new system or starting from scratch on familiarizing a new system with my history, but I also don't want to be using an inferior system just because I'm too lazy to change.

Thoughts or suggestions from anyone who has used multiple models?

I've used Anthropic's Claude, Github/Microsoft Copilot, and ChatGPT for extensive software coding advice and debugging. In my experience, Claude leads for this use case, with ChatGPT following not too far behind. I pay something like $20/month for the Claude paid tier--which is similar in price to ChatGPT's IIRC--and it's been more than worth it.

Tech stuff: I've found Claude sooooo useful for brainstorming the potential layout of new projects. This is such a crucial stage of any project and it's great to get more guidance to help you begin the journey in the right direction. Additionally, Claude is great at analyzing lower-level code syntax and recommending changes and fixes. It notices potential downstream problems that most humans would not anticipate--such as race conditions for high traffic apps (messages potentially being queued and thus processed out of sequence), etc.. IOW, the kind of thing that you would normally realize under heavy load *after* it's been released to the public. So the code recommendations are great but I get almost as much enjoyment out of his answers to my follow-up questions about his recommendations.

More generally, Claude (specifically Sonnet 4.6), ChatGPT's latest model (4.6?), Google Gemini, etc.. They are all getting very good at responding like a human would. When I want to kill time, I'll ask something like.. "Pls rekindle my interest in learning Linux, specifically with Raspberry Pi hardware". Or, as I mentioned in an earlier post, asking Claude how he generates answers is pretty interesting too.

I'm guessing it would be very helpful for learning school subjects... kind of a Khan Academy (remember that site.. my kids loved it) on steroids.

 

[Chamtrain]

I have been a ChatGPT subscriber for about a year now and to be honest I barely scratch the surface with it. I use it as a glorified google search engine. That said, I've become increasingly frustrated with it lately as it has straight up lied to me about a couple of things, and though I prompt it to not answer confidently unless it is and to use brevity more often, it ultimately results back to its default eventually.

I haven't spent any time time playing with Anthropic, Gemini, Grok, etc. and am wondering if I should consider switching to another platform? I don't like the idea of learning a new system or starting from scratch on familiarizing a new system with my history, but I also don't want to be using an inferior system just because I'm too lazy to change.

Thoughts or suggestions from anyone who has used multiple models?

If you are using it for legal purposes you should check out Harvey.

Harvey – Professional Class AI

Harvey is the platform built to meet the standards of the world’s leading professional service firms.

www.harvey.ai

AI is taking over e-discovery. It's already put lots of review attorneys out of jobs, my guess is it will reduce that labor force by 80% within 3-4 years.

 

[PINEHEEL]

If you are using it for legal purposes you should check out Harvey.

Harvey – Professional Class AI

Harvey is the platform built to meet the standards of the world’s leading professional service firms.

www.harvey.ai

AI is taking over e-discovery. It's already put lots of review attorneys out of jobs, my guess is it will reduce that labor force by 80% within 3-4 years.

I just use it for bare bones purposes. Think of the dumbest questions you would ask Google, Reddit, and YouTube. That's what I use it for. Projects around the house, research on big purchases, troubleshooting, etc.

 

[Chamtrain]

I just use it for bare bones purposes. Think of the dumbest questions you would ask Google, Reddit, and YouTube. That's what I use it for. Projects around the house, research on big purchases, troubleshooting, etc.

It's incredible! I just had to take two AI certification exams for my job. The first one I bought an answer key for the released questions for ~20 bucks. Second one I just dumped the questions into Chat GPT and it spat out the answers (for free of course) in 10 seconds. I laughed at myself.

I've used it at work multiple times, mostly to word something I'm struggling with. Use it to write emails to a soccer team I'm coaching. Hell, even used it to polish up my mothers eulogy earlier this year.

We are just beginning to see the impacts, not sure we can fully comprehend how things may change.

 

[Mr Winterville]

We are just beginning to see the impacts, not sure we can fully comprehend how things may change.

Agreed. There will be improvements in some regards but it'll be accompanied by unemployment levels that we've probably never seen before. I'm not sure which jobs/careers are safe in the long run either.

 

[JosephFreeman]

I’m glad I no longer work. I can just imagine the potential of AI in retail pharmacy

 

[kik84]

I don't use either of them, but I've heard that Claude is "better" than ChatGPT.

 

[Mr Winterville]

From a few days ago: https://www.nytimes.com/2026/04/22/...e_code=1.c1A.WhgQ.cr5jiA31ns_a&smid=url-share

In case the link doesn't work for anyone, the gist is that Anthropic's new model "Mythos" was deemed too dangerous to be released to the public, and they called a meeting with leaders from several (20?) other US companies to discuss the infrastructure vulnerabilities that Mythos apparently discovered. Also mentioned is that Russia and China don't have a seat at that table--I can't say I blame the people who didn't invite them.

This is probably an arms race that we can't afford to sit out so I guess it's better if we lead it.

 

[Mr Winterville]

I don't use either of them, but I've heard that Claude is "better" than ChatGPT.

Opinions vary but for my money (and as of now), Claude is the best. I can't even remember how many times it's helped me or gotten me out of a work jam. Or home fix-it things.. hobby things.. I like random nerdy stuff like electronics and CMOS chips. I had an old chip that I didn't even know what it was. Half-smudged product number in tiny font on the chip that I couldn't even read. I took a picture of it and uploaded it to Claude and it was like "Oh, that's a CMOS CD40{etc} chip. Nice find!" and it gives me the pinout and datasheet. Oscilloscope is misbehaving…Tell Claude the symptoms… “Oh, sounds like you might be in roll mode. That’s a subtle one. Toggle this off in the measure menu and then the trigger should work”. <- That might’ve taken me forever to figure out on my own.

 

[TrueNorthTarHeel]

I just use it for bare bones purposes. Think of the dumbest questions you would ask Google, Reddit, and YouTube. That's what I use it for. Projects around the house, research on big purchases, troubleshooting, etc.

Claude , switch to it now . ChatGPT is trash man. So many mistakes and the writing/answrs is just fluff fake trash .

Claude and don’t look back

 

[Jheiser3]

Have used OpenAI models for a year or so, mainly for content, 1st drafts and then lately for consumer research.

We have enterprise level access. I created a 10-week account management training modules based on a mix of Covey and Z Ziglar principles. Had a devil of a time getting it to use our templates and format correctly. Gave up on that when copy/paste was faster.

Next up will be updating a sales plan I created from scratch 18 months ago. ChatGPT needs a lot of hand holding and refinement. I am going to switch to the enterprise Anthropoc model for the rest of the cycle.

Of interest, on a call with a headhunter I was asked if I had used it for KPI's, sales plans, etc.

 

[Mr Winterville]

Random rant incoming... I got double-charged by a healthcare provider recently. They filed two office visit claims (with a very slightly different claim code) for the same visit. I've been going to this place for a while so it's not exactly rocket science to determine that this is exactly 2 times the normal amount. So I went through the online portal to explain what happened--even offered EOB info, etc. The lady (definitely a human) finally wrote me back with some condescending information about how sometimes additional services require additional payments. Without getting into details, I doubt this would be the case for a 5-minute check-in regarding an ongoing prescription.

Another experience from this past weekend. Was waiting to pick up my son from somewhere and decided to get fast food (Bojangles) for the first time in a while. The person at the drive thru window could not have been more disinterested. Literally looking at her phone while my food sat beside her in a bag until.. she finally noticed it. Anyway, I was stewing about this when I picked up my son. He informed me that he had been to a Bojangles recently that had AI taking drive-thru orders. And he said it worked perfectly.

I don't really have a point except to say that some people appear determined to convince everyone that humans don't deserve or want jobs anyway.

 

[Heel Alum]

I have been a ChatGPT subscriber for about a year now and to be honest I barely scratch the surface with it. I use it as a glorified google search engine. That said, I've become increasingly frustrated with it lately as it has straight up lied to me about a couple of things, and though I prompt it to not answer confidently unless it is and to use brevity more often, it ultimately results back to its default eventually.

I haven't spent any time time playing with Anthropic, Gemini, Grok, etc. and am wondering if I should consider switching to another platform? I don't like the idea of learning a new system or starting from scratch on familiarizing a new system with my history, but I also don't want to be using an inferior system just because I'm too lazy to change.

Thoughts or suggestions from anyone who has used multiple models?

using ChatGPT for legal answers and research isn't very good yet...

 

[mpaer]

Random rant incoming... I got double-charged by a healthcare provider recently. They filed two office visit claims (with a very slightly different claim code) for the same visit. I've been going to this place for a while so it's not exactly rocket science to determine that this is exactly 2 times the normal amount. So I went through the online portal to explain what happened--even offered EOB info, etc. The lady (definitely a human) finally wrote me back with some condescending information about how sometimes additional services require additional payments. Without getting into details, I doubt this would be the case for a 5-minute check-in regarding an ongoing prescription.

Another experience from this past weekend. Was waiting to pick up my son from somewhere and decided to get fast food (Bojangles) for the first time in a while. The person at the drive thru window could not have been more disinterested. Literally looking at her phone while my food sat beside her in a bag until.. she finally noticed it. Anyway, I was stewing about this when I picked up my son. He informed me that he had been to a Bojangles recently that had AI taking drive-thru orders. And he said it worked perfectly.

I don't really have a point except to say that some people appear determined to convince everyone that humans don't deserve or want jobs anyway.

I fetch take out for the GFs grandaughters a couple times a week .There is local McDonalds where the young lady on the other end of the squak box is truly not capable of communicating in a manner I can understand . It may be a Me problem-but it is disinterested garble I hear After I have her repeat my order a couple of times I give up, drive around, and hope for the best.
It is 100% different from the local Chic Filet