I’ve noticed over decades of giving IT support and teaching people how to make the best out of their computers, that most are inefficient when googling. They lack the smarts to make a short, to the point, query.

AI is a complete new game.
How about sharing your tips @ tricks?

Here’s the ones I’m using. Self explanatory.

- Detailed
- Step by step
- Printable

For the time being, im using Google. I’m aware of Claude, Grok and others but have no need for them ATM. Feel free to share why you find yours more interesting or practical.

Exploring the practical use of AI, I just toyed with my Man Cave’s audio equipment. Provided a list of hardware components, software and asked how to best combine them. Impressed how AI handled the task, using manual and specs of very old stuff and current audiophiles shared information.


0-0

I have found that Claude is much better at technical questions - especially related to engineering or IT...

I've only been using as a resource for 6 months or so, but have improved my ability to ask very detailed questions and to include multiple levels or details I want answered all in one query.

To me this is the help in these tools over doing an old web search then drilling down to more and more detail on my own. I know what my second/third/etc. level questions are going to be. It takes a little longer to fashion that detailed of a query but it gets to the answer much more easily.

Especially important if you are using free versions.

Like OP I've used Google some after really getting started with ChatGPT. I find GPT to be more prone to incomplete or inaccurate answers that I have to redirect. I have lately used Claude, and find it to be MUCH better for technical or very analytical queries.

Related to this, if you are using free versions be sure to check which version/engine you have selected as default. For example in Claude using "Sonnet" will get you way more usage time than if you are toggled to "Haiku" though the latter is faster and perhaps more refined. The processing allowance is less with the latter however. Similar rules in GPT and I assume others.

As a recent documentary noted, all current AI models are really doing is processing for "the best next word" to include in response. Thus the LLM moniker. Thought of this way it will help you understand that you are really using an incredibly powerful search engine, composer and dictionary, not an actual thinking intelligent tool. This may help explain and set expectations for new users as well.

That level of intelligence is a long ways off.

I have found that uploading photos to AI applications is very helpful. For example, I was in a hotel room and was having a problem mirror casting a movie I had downloaded on my iPad onto the hotel room's TV. Taking photos of the error screens on the tv and uploading them to Claude allowed the AI to see what was going on and provide quick solutions.

Google AI often cites Reddit, Facebook, etc as the source. Less than reliable sources IMO. Super Grok appears to be doing fundamental research and provides concise, comprehensive, and well-organized responses. Pricey, but worth it to me.

Claude can work with your files in VS Code and Obsidisn.

Markdown files in either, you can ask Claude to factcheck your writing and fix grammar. He loves em dashes though. They are correct but it overdoes it. I hook in a humanizing pass to get rid of them.

quote:

Originally posted by Georgeair:
… For example in Claude using "Sonnet" will get you way more usage time than if you are toggled to "Haiku" though the latter is faster and perhaps more refined. The processing allowance is less with the latter however. …

Haiku is the cheaper model. Then Sonnet, Opus and then Fable at the top.

Very glad to see you posting again Mark!

I am a fan of Perplexity! I tested several others and found them wanting… not so Perplexity. I require original sourcing for answers, not news or republished information.

So far, I have not been disappointed!

quote:

Originally posted by mark123:
Haiku is the cheaper model. Then Sonnet, Opus and then Fable at the top.

My mistake. When I first went to Claude I decided to split the difference and use Sonnet, forgot it was a step down from Opus, not Haiku. On Opus it timed out for a half day on my first query due to using up my limit!

quote:

Originally posted by Georgeair:

quote:

Originally posted by mark123:
Haiku is the cheaper model. Then Sonnet, Opus and then Fable at the top.

My mistake. When I first went to Claude I decided to split the difference and use Sonnet, forgot it was a step down from Opus, not Haiku. On Opus it timed out for a half day on my first query due to using up my limit!

I ran Fable on the $20 tier when it was available without usage credits. A code review ate the weekly limit up rather quickly.

“Oh, you have a free weekend? Great. See you on Tuesday.”

Here are some AI tips and tricks I routinely use:

I'm a gardener and there is so much bad advice out there from old wives tales, people parroting the bad advice they heard, stuff published to generate clicks, etc. I specifically ask for the response to be based on university research or university extension publications.

Gardening and lawn care advice is contextual such as cool season lawns (i.e. northern US) vs warm season lawns (i.e. Southern US), herbs and vegetables (e.g. tomatoes) grown in hot, humid Gulf coast where we have 4 months of misery vs advice from a place that has 4 days of that misery. I don't want to follow some advice like using an herbicide that successfully works on rye grass, kentucky blue grass, and tall fescue, but would kill St. Augustine grass as it'd cost me thousands to resod my lawn. I specifically ask for advice based on specific plant/grass only from university research or university extension publications from states in the Gulf Coast.

When I'm looking for data, I get so damn sick of it giving links to Facebook, Reddit, YouTube and other social media outlets which are polluted with idiocy. I specifically ask for data from original data sources not Facebook, Reddit, YouTube and other social media outlets.

I actually click the links, and challenge if the link doesn't match AI's assertion. For example, I used Grok AI yesterday and it told me how to modify something I already own rather than buy the expensive current option that already has that feature. However, when I clicked on the links they were all for the expensive current option not modifying what I already own. I called bullshit and it provided helpful links.

I actually read the data it publishes and challenge if the data doesn't match AI's assertion. For example, I used Google's Gemini AI this afternoon and it tried to tell me Prison Ministry Fellowship (i.e. the organization Chuck Colson founded) doesn't outperform every federal recidivism program then posts data from 4 studies showing it was 40x to 60x better. I called bullshit and it apologized and admitted the its conclusion was erroneous and they did outperform everything including the one, First Step Act, that Kamala voted for as a Senator.

---

Ego is the anesthesia that deadens the pain of stupidity

DISCLAIMER: These are the author's own personal views and do not represent the views of the author's employer.

quote:

Originally posted by tatortodd:
I called bullshit and it provided helpful links.

Thanks for your comments, and that one is especially important. Pushing back very often yields significantly different results.

My experience with AI has been that its knowledge is very incomplete. It frequently offers inaccurate information. But it will correct that information if you challenge it. But of course that is not very valuable if you are asking for information about which you are unfamiliar.

quote:

Originally posted by tatortodd:

When I'm looking for data, I get so damn sick of it giving links to Facebook, Reddit, YouTube and other social media outlets which are polluted with idiocy. I specifically ask for data from original data sources not Facebook, Reddit, YouTube and other social media outlets.

I like to use the phrase "primary credible sources."

Google Gemini sucks. It will just make shit up; gas light you; or ignore explicit instructions. I got a new phone and it came with six months free advanced model. I'm thinking about cancelling my FREE subscription.

Another problem with the different versions of AI is that, when you start to dig deeper into an issue that seems to be leading in a certain direction, AI starts to advocate for the mainstream media position. I had to tell it more than once to just answer the questions and to stop advocating for any position.