Sunday, I was speaking with my nephew, who swears by using AI for just about everything. Explaining to him of my frustrations of using Google without success. I was looking for very specific work related topic for about 10 years. My peers were of no use. I couldn't find this info at the Public Library. He strongly suggested ChatGPT.

Fast forward to yesterday, my first time trying it, thinking the worst. To my utter surprise, the results went from almost nothing on Google to extremely detailed & complete info from ChatGPT!

It appears that I might be using this more than Goolge going forward.
Since I'm new to using AI, are there better choices to consider?

ChatGPT and Gemini are the two I use frequently. You'll get a feel for what each does better.

Remember that they hallucinate and will make up things or miss very obvious data points that Google doesn’t. I would only use them as part of your process and never solely rely on them for anything that is important.

I've lost count of how many times ChatGPT has been flat-out wrong or has provided crappy advice.

Just today, I was looking to replace a WordPress plugin that had a security flaw and was removed from the repository YEARS ago. For some reason, it's still available online.

When I asked ChatGPT what I was trying to do - defining the functionality requirements I needed for the site - the first "clean, simple and most flexible" option was the plugin I was trying to replace!

I responded, "The first option you suggest - This plugin has been closed as of October 27, 2023 and is not available for download. Reason: Security Issue."

ChatGPT response... "Yep — you’re right to avoid it. “[Plugin Name]” was closed due to a security issue (it shows up in vulnerability reporting for that timeframe), so it’s not something I’d recommend installing."

But for some reason, it suggested it as the first option?

This is a daily occurrence for me, so your nephew and everyone should be aware.

Without a doubt, ChatGPT and others have saved me time on many things, including helping me with a bit of PHP and CSS coding here and there. Enough so that I pay the $20 per month for a ChatGPT subscription.

But you REALLY have to check the work product or suggestions they provide.

I like Claude - it is supposed to be geared towards problem solving. It gives me better answers when compared side by side with the others.

Grok is the only AI with a mandate to be "Maximally truth seeking"

and so Grok is the only AI I interact with.

ChatGPT has been a terrific help in my professional life of late.

It has served well in educating the people I work for in all the things we've not done properly and should revisit to achieve the absolute truth presented by the AI.

I admittedly struggle with math. Algebra and simplifying are especially a weakness. So, I used Google Lens to help me out. I, too, have noticed that it isn't reliable. In one instance, it took about 10 attempts to finally get the correct answer.

Like you folks mentioned, I warned my nephew to check all that it spits out.

That is part of the reason for asking what might be better alternatives.

ChatGPT seems mostly good for facts and Grok seems a bit better for recent events and opinions. Both can seem correct and be completely wrong. Sometimes it helps to post the same request to both or copy the results from one and ask what the other thinks of them.

Despite working in IT, I rarely, if ever, use AI. My experience has been that AI is nothing more than a data aggregator. It doesn't provide anything "new". It just aggregates all of the data on the web. If some of the data is wrong, it will present the incorrect (wrong) data.

I'm not saying AI doesn't provide value, I just think it's over-hyped and actually does less than what is claimed.

I use ChatGPT (I have a lifetime subscription on an app I signed on for just $45). It has since given me access to all other AIs.

I use any AI for just about any purpose, mostly reviewing what I want to write for spelling, punctuation, grammar, and readability. Sometimes, I have to direct it to keep my "voice" in its suggested rewrites. Any will do for that and even for "light" research like "I have a song running through my head and the only line I remember is something about stars. It was a popular song maybe 10 years ago." That started a back and forth with Google Gemini that finally got to the actual song: Counting Stars by OneRepublic.

For serious technical stuff, I use Grok and I pay $300 a year for the subscription. Well worth it as my main purpose is to help me manage my portfolio.

But like any tool, you have to know how to use it to be effective and it's been a learning journey for me as well.

For example, my prompt for it to do a stock market analysis is 1500 words that it even helped me to create. It started from a simple query "What indicators should I monitor to analyze where the stock market is going?" I got it to a robust status where the prompt's purpose is to give allocation changes to a baseline I feed it which was its last analysis for two portfolios. The goal is to optimize near-term (3-6 months) returns mitigating 20-30% downturns for two portfolios, 1 being capital preservation oriented and 2 being growth-oriented. One time, I did ask it to evaluate the prompt if the prompt directs it to overweight sectors that will take advantage of any bullish trends and its answer was 'No, you only directed me to modify the prompt to mitigate 20-30% downturns in the near-term. That started another round of tweaks where it now looks for optimization strategies for recovery/bull periods.

Especially for math problems, you have to have an idea of what the reasonable range of the result should be. Just recently, I asked it for help in giving me the net present values of increasing cash flows to model in Excel. The numbers didn't look right to me. I was right that the results were off but the error was attributable to me missing part of the formula it gave me to enter for each cell. But other times, I did catch AI being incorrect or "biased" in a particular way, especially on some social issues (This was with Google Gemini). I had to say, "Isn't such and such the right answer?" And it said, "You're absolutely correct in catching that error!" or "Yes, some other sources say such and such...."

quote:

Originally posted by bryan11:
...Sometimes it helps to post the same request to both or copy the results from one and ask what the other thinks of them.

That is an interesting suggestion.

I have noticed that both Google and ChatGPT struggle with "tangent" when I describe the geometry via text or pictures. They both seem to answer towards an intersection point instead.

Claude AI for me. $20 per month. Far better user interface and much better software coding responses.

I'm still "AI" averse - I haven't used any AI app to my knowledge. I still depend on search engines. I am conceptually familiar with AI data sources like web crawling and LLM.

If you ask an AI app about the data sources used for its response and its confidence in the accuracy of the response, what would it say? Would it provide a detailed assessment of how it developed a particular response? Or would it be just generally conceptual?

I just tried it. It said my car battery was made in 2015. It was wrong. It was made in 2021.

Using Claude & Grok. Using both, keep each other honest.

Gemini -Google, was a terrible experience. Obvious influence and bad sources.

I've caught ALL of them making up sources. You have to check the references and citations, and input restrictive parameters that clearly instruct them NOT to make up data or sources.

I've been redoing our WP based website, and both have been invaluable for helping me clean up code. A work in progress, but helpful. I've also been using both for research. They've found some sources I wouldn't have found without help, and overall are more help than hurt.

Don't trust them, though. None of them.

Perplexity is another.

About eight months ago, I started a thread about using two different AI (Grok and Meta's) to create a dry rub wing recipe, and each took a completely different route. I still use the Grok recipe and last used it two days ago. I didn't try ChatGPT.

Earlier this week, I used ChatGPT to create the caricature for the Sigform thread. It did a nice job. However, I tried to make one for my buddy and it completely screwed it up then proceeded to repeatedly screw up my instructions for correcting the error. I ended up hitting the limit for maximum number requests before it was successful, and I didn't bother to come back an hour later (i.e. after the lockout period) and try again.

I rarely google anymore, as it gets so frustrating to have their advertisers pushed first in the results and it sometimes the third page before they give a legit response. I have found that Grok, Gemini AI, and Microsoft CoPilot all produce better search results then Google.

quote:

Originally posted by tatortodd:...proceeded to repeatedly screw up my instructions for correcting the error...

They all struggle with images. Particularly changing ones that already were created. They don't apply new instructions to current images, they redraw every image pixel by pixel every time.

At least that is what both Grok and Chat explained to me.