AI being helpful thread

[bmaverick]

I asked 3 different AIs on how to DIY project with items from around the garage and hardware store.

What I mostly received in the reply is the history of how a VAWT was made and used without any real instructions.

I found a 4th AI off the beaten path, LUMO at Lumo: Privacy-first AI assistant where chats stay confidential . I typed in the same 5 words in the same order, ... DYI make vertical wind turbine ... Received a wealth of info, step-by-step instructions, sources on how to obtain common parts already in the garage and to use common tools. And lastly, it did have improvements if I would consider owning and learning a little in 3D modeling and 3D printing to make this hobby into a side line retirement income to make and sell to others. The geometry has a higher efficiency for the single home owner vs a utility company.

 

[JeepHead]

You get out of AI what you put into it. How you ask the questions determines the answers you will receive. The more specific you are, the more it is. If you tell it to ignore opinions and to only give credible, factual information, that's what you will usually receive. You can even have it give you the sources for the information it's presenting to you so that you can confirm the accuracy. AI is becoming better with each day and the more people use it, the faster it will learn. AI will grow at a rate faster than we have ever seen anything grow before. It's here to stay.

Once you learn how to write correct prompts, it becomes much more powerful.

 

[scootr]

I asked ChatGPT how to get the best answers:

1. Be specific about the task​

Vague:

“Tell me about heat pump water heaters.”

Better:

“Explain how a heat pump water heater works, then compare its annual operating cost vs a 40-gal propane water heater for a 2-person home in California. Assume current 2025 energy prices and show your assumptions.”

Specify:

• Goal (compare, explain, design, troubleshoot, summarize, etc.)
• Scenario (home size, budget, experience level)
• Output form (table, bullets, step-by-step, code, etc.)

 

[garry1p]

I don't think anyone "programs" AI to give biased or false information.
AI may provide questionable or on rare occasions wrong information however, it is "programed" to continually improve on the bases of new information.
I also think when AI provides the wrong or questionable information it is based on how the information was asked.

 

[3930dave]

It's important to realize that AI isn't programmed to deliver the truth, it's programmed to deliver the answer that its programmer wants it to give.

Bias - it's what us kids were usually taught to examine, back in the Olden Days. Before reading a book in English Lit, you were expected to understand what the author's background (social, economic, education) and upbringing were.

I happened to already really like to read, before English Lit, but the course would have been valuable if that ^ was the only thing I learned from it.

I do actually appreciate the physical-project related examples people have posted here (I haven't had AI playtime to spend....), as they are relevant low level examples of what this tech can do.

The longer, high-level trajectory is what concerns me. Mega corps making trillions off of herding us in a certain direction is bad enough.... what happens when AI is totally off-the-leash, if it's not there already, In The Wild ?

Meaning, when no human programmer (Good/Bad/Indifferent re. Ethics, Morals.....) is involved at all, and for The Rest of Time ?

Query - Given all the Useless Eaters now out of work, AND, that Humans are a Plague on The Planet, what is the most efficient solution to this problem ?

Powerful Tech, can have Powerful Results...... esp. when No Guardrails or Leash are present.....

Rgds, D.

 

[WinterDeere]

I don't think anyone "programs" AI to give biased or false information.

If you had to summarize it in one sentence, that's sort of what sets AI apart from traditional programming: "no one programs it"!

One interesting demonstration of this came about in 2017, when (once again) using chess to demonstrate the new AlphaZero AI model. They had the AI play against itself, having already been exposed to all of the classic strategies, and it completed something like 10,000 games overnight. The next day, those involved in the project saw the AI executing strategies that just made no sense at all.

Initially, they thought the learning model was broken, but then the realized it had invented totally new and novel strategies that no chess master had ever even considered before. Again, not "programmed", but "learned".

 

[quicksandfarmer]

I don't think anyone "programs" AI to give biased or false information.
AI may provide questionable or on rare occasions wrong information however, it is "programed" to continually improve on the bases of new information.
I also think when AI provides the wrong or questionable information it is based on how the information was asked.

AI is trained. It's presented with questions and the answers it gives are compared against known answers to those questions. The scoring is then fed back and it "learns" to give correct answers.

While the AI trains itself, the training sets that are used are selected by the designer. If they want to steer responses in a certain direction they can select training sets to do that.

 

[paulsharvey]

One limit ive found, AI doesnt have access to some old media info. I mean newspapers, industry magazines, ect, where some very specific info would be found. Some of it is scanned and available online, if you know where to dig. That will improve over time likely, but it IS limited by what's actually out there in E form.

Ive also caught Deepseek giving good sounding, but false, historic info, based on it not finding records.

 

[3930dave]

Ive also caught Deepseek giving good sounding, but false, historic info, based on it not finding records.

So it's already exhibiting "normal" human behaviour.....

It was only a matter of time .

Rgds, D.

 

[paulsharvey]

So it's already exhibiting "normal" human behaviour.....

It was only a matter of time .

Rgds, D.

So, the specific example, I gave in another AI topic thread, but it was about an iron casting foundry. I only looked into after seeing a Very old manhole lid, with the foundry name. So, with Google searching, I found a mention of it into the 1930s, it was a picture from a trade magazine.

Anyways, Deepseek, said the foundry was destroyed in 1864 by the federal raids when they burnt the city. Not true, but with "deep think" on, it came to that conclusion by the lack of information later, and that a foundry would be a target when the union attacked. So, it took the lack of available into as proof that it was destroyed.

Do, it spit out a good sounding story about it beeing destroyed in Feb of 1864, can the commanding officers name, unit info, ect. Sounded good, just not true