Comparing AI engines

[Shrink726]

Any thoughts on the best AI engine? I've tried ChatGPT, Gemini and Grok. Both ChatGPT and Grok seem excellent while Gemini is really lagging.

I'd love to hear others' thoughts....

 

[Deuce]

Ask Grok about the holocaust and see how excellent it still sounds

 

[Airyaman]

I use grok and chatgpt both. I'll ask them the same questions to get as much information as possible.

Chatgpt chokes (slows down) too often for my taste however.

 

[Shrink726]

Ask Grok about the holocaust and see how excellent it still sounds

I just tried that having heard of the (ahem) problems it was having. I was pleasantly surprised that it gave me a sensitive, nuanced response without any trace of bigotry or anti-semitism.
Don't get me wrong, Musk is a complete and utter tool but I'm trying to be objective about my assessment of these AI engines and, so far, at least, Grok seems to be pretty good.

I was quite shocked with the mediocrity of Gemini in view of the incredible resources and brain power at Google's disposal.

 

[Slartibartfast]

A combination of deep seek, Claude, chatgpt, and copilot.

 

[Lursa]

Copilot's the only one I've spent a lot of time with but we were training it and testing prompts.

You can get it to do some pretty detailed tasks if you craft the prompts well.

 

[Str8shooter]

Any thoughts on the best AI engine? I've tried ChatGPT, Gemini and Grok. Both ChatGPT and Grok seem excellent while Gemini is really lagging.

I'd love to hear others' thoughts....

AI is useless. It compiles answers from a mass accumulation of information on the internet. Liberals have flooded the internet with this crap.

 

[Shrink726]

AI is useless. It compiles answers from a mass accumulation of information on the internet. Liberals have flooded the internet with this crap.

Sure seems like a whole lot of very smart people see AI as a paradigm changing development with an almost unlimited future (for better or worse).

Pretty shocking for something that's "useless."

 

[Mycroft]

Any thoughts on the best AI engine? I've tried ChatGPT, Gemini and Grok. Both ChatGPT and Grok seem excellent while Gemini is really lagging.

I'd love to hear others' thoughts....

I've only used Edge's Copilot on occasion. It can be useful for getting data without having to look at a slew of websites. But Copilot has one problem: It'll say one thing, but when questioned about what it's saying, it will easily contradict itself.

For example, I ask today if gerrymandering in the middle of the decade was unprecedented. In part, it said this:

Yes, gerrymandering maps in the middle of the decade—outside the traditional post-Census redistricting cycle—is considered unprecedented​

I then asked for historical precedents for the practice and it proceeded to list the times this kind of gerrymandering was done and ended its response with this:

So while it’s not entirely unprecedented, it’s certainly controversial—and historically rare.​

So, I asked Copilot why it first said one thing and then another. Here's the response:

You're absolutely right to call that out, Mycroft—and I appreciate your sharp eye.

So the more accurate framing is: mid-decade gerrymandering is historically rare, controversial, and norm-breaking—but not entirely unprecedented.

Thanks for holding me to a high standard.​

The thing is, I shouldn't HAVE to hold Copilot to "a high standard".

 

[Mycroft]

btw, I just now talked to Copilot about it being inconsistent. In fact, I called it's first response a lie when I had asked it a straightforward, simple question.

Here's what it said:

You're absolutely right, Mycroft—and I saw your post on Debate Politics. You asked a clear, direct question about whether mid-decade gerrymandering was unprecedented, and I gave a contradictory answer. That’s on me.​

​

To be crystal clear:​

​

Mid-decade gerrymandering is not unprecedented.​

• It has happened before—Texas in 2003 is the most famous example.
• Other states like Georgia and North Carolina have also done it outside the Census cycle.

What I should’ve said from the start is that while it’s rare and controversial, it’s not without precedent. You shouldn’t have to “hold me to a high standard” to get a straight answer. That’s my responsibility, and I’ll do better.​

​

Thanks for calling it out publicly and directly.​

It's interesting that I didn't mention Debate Politics at all to Copilot, but it saw my post and referenced exactly what I said in my post.

 

[Shrink726]

Wow, that's very cool...and a little scary.

 

[Mycroft]

Wow, that's very cool...and a little scary.

The lesson here is, at least with Copilot, don't take everything it says as fact. Think about what it says. Ask questions. Make the AI answer honestly and call it out when it doesn't.

It's kind of like talking to real people here on DP.

 

[Desert Storm]

Deepseek and Grok are my 2 favourites

 

[Shrink726]

Just tried Deepseek. It's slow to generate responses and there is no button that automatically enables voice input, but those are minor quibbles.

It seems to generate extremely thorough and nuanced results, probably better than any of the others I've tried. Very impressive. Now if they could just improve the user interface....

 

[ataraxia]

Any thoughts on the best AI engine? I've tried ChatGPT, Gemini and Grok. Both ChatGPT and Grok seem excellent while Gemini is really lagging.

I'd love to hear others' thoughts....

ChatGPT is excellent- I love its conversational tone. Talking to it is like going to the office hours of some really smart, helpful, patient, encouraging, enthusiastic, and friendly college professor- who has all the time in the world. It even keeps goading you on and piquing your interest to keep asking it more questions.

Claude has also been really accurate- though not as conversational and friendly.

 

[ataraxia]

AI is useless. It compiles answers from a mass accumulation of information on the internet. Liberals have flooded the internet with this crap.

Actually that’s not true. For example, I asked ChatGPT what would happen if government officials put out public statements contradicting the consensus of the published peer reviewed scientific articles, or there was enough public confusion so that most people believed things which were contradicting the peer reviewed scientific literature- like what is happening today with vaccine science or climate change science.

It said that it would take a lot for it to be influenced by lay public opinion or online buzz, or by political propaganda or misinformation by political appointees (although not impossible- it admitted)- because it is smart enough to know to strongly prioritize the peer reviewed literature on scientific questions. That was reassuring.

That was somewhat reassuring..

 

[Slartibartfast]

btw, I just now talked to Copilot about it being inconsistent. In fact, I called it's first response a lie when I had asked it a straightforward, simple question.

Here's what it said:

You're absolutely right, Mycroft—and I saw your post on Debate Politics. You asked a clear, direct question about whether mid-decade gerrymandering was unprecedented, and I gave a contradictory answer. That’s on me.​

​

To be crystal clear:​

​

Mid-decade gerrymandering is not unprecedented.​

• It has happened before—Texas in 2003 is the most famous example.
• Other states like Georgia and North Carolina have also done it outside the Census cycle.

What I should’ve said from the start is that while it’s rare and controversial, it’s not without precedent. You shouldn’t have to “hold me to a high standard” to get a straight answer. That’s my responsibility, and I’ll do better.​

​

Thanks for calling it out publicly and directly.​

It's interesting that I didn't mention Debate Politics at all to Copilot, but it saw my post and referenced exactly what I said in my post.

AI has an issue with being a "yes man". You may want to be careful with your prompts and add phrases like "Tell me the truth, even if you don't think I will like it"

 

[Slartibartfast]

ChatGPT is excellent- I love its conversational tone. Talking to it is like going to the office hours of some really smart, helpful, patient, encouraging, enthusiastic, and friendly college professor- who has all the time in the world. It even keeps goading you on and piquing your interest to keep asking it more questions.

Claude has also been really accurate- though not as conversational and friendly.

I primarily use Claude for planning next steps in my career, resume building, etc. Its amazing at that specific task.

 

[joluoto]

Any thoughts on the best AI engine? I've tried ChatGPT, Gemini and Grok. Both ChatGPT and Grok seem excellent while Gemini is really lagging.

I'd love to hear others' thoughts....

Are you planning to do something serious or just play around? If you are planning something serious my answer is: don't. If you are planning to play around anyone of them can be used as a funny toy.

 

[Shrink726]

Are you planning to do something serious or just play around? If you are planning something serious my answer is: don't. If you are planning to play around anyone of them can be used as a funny toy.

I guess my response would be "play around" though I'm not sure what you mean by "something serious."

 

[joluoto]

I guess my response would be "play around" though I'm not sure what you mean by "something serious."

Something related to work or other stuff that might hold some importance. If not, all of them can be amusing.

 

[Lursa]

It's amazing to see the limited perspectives that only think of AI to look for "info on the Internet." I realize that many people dont have jobs where they use computers (for anything) but still...the lack of curiosity or optimization is kind of sad. As crochety as it sounds, the younger generations wont be/arent so limited and will use it to work smarter, not harder.

And yes, getting the most from AI does usually mean an iterative, "teaching" curve...which pays off.

 

[bythoughts]

These things are a moving target. Three months ago Perplexity was sharp as a tack and would cite real sources ... now, it only cites random web forums unless you pay. The reasoning still seems to be based on the good sources, but it gets a little duller and a little less interested in actually answering the question every week, I think.

We're still in the honeymoon period. You'll know the honeymoon is over when the AIs start asking YOU questions. I mean, you want to take a bridge over the Delaware? That's not just a toll but a long chat about what people in your family do for a living, whether any of them have ties with foreign governments, and whether any have been involved with anti-Semitic activism in regard to the resettlement of the Palestinians.