I Determine How Much Ai Understands About Gold Detecting

[jasong]

12 minutes ago, Clay Diggins said:

I think if we can get rid of this nonsense idea that AI actually exists today (it doesn't) we might better understand these ongoing battles to see who will own your personal data and browsing habits in the future.

 

I disagree. AI exists. AI existed when I was still in college, my roomate was doing grad work on neural nets and that was my first exposure to it. Even the crude models back then demonstrated some kind of thinking. What they didn't (and probably still don't do) is consciouness. But there is some kind of reasoning happening with these current models of GPT. They can solve novel problems, and they can find new ways to solve unsolved problems. They can create brand new things on their own, even without user prompts. I get that some don't see the power in this, but I do, and to me the semantic argument over what "intelligence" is or isn't is a moot point. That to me, is intelligence.

What doesn't exist is AGI. And this will, IMO, be the point of discussion in coming years.

Also, those articles mostly years old. I can't over emphasize how much has occured in just the last 6 months in terms of development. Literally, people are starting to talk about AGI potentially being here, or close now.

This stuff is growing exponentially fast. I've never seen anything like this since the dotcom days. 

[jasong]

Ok well right when I was sure it was lying to me about really analyzing photos/visual data, I tried another experiment and now I'm not so sure. 

I gave it the weird rock posted a week ago in the rock ID section. And it definitely is actually analyzing pictures, and it's scraping up to date web info even though it says OpenAI rules prevent it.

I set it on an iterative loop where it was simulatenously running a few instances of AI's, all of which came to different conclusions at first oddly (basalt lava flow, granite, desert rose, sandstone, etc) but ended up comes to the conclusion that it's banded chert with an iron oxide coating. Of which I think I agree. But it says it isn't possible to provide an accurate ID without further in person tests - also agree.

It still hasn't eliminated sandstone, jasper, granite with texturing (?) and some kind of fossil(s) though, but it seems to be stuck and is requiring human input for testing to go further so I stopped it. Of interest - it didn't mention a concretion - which it may actually be and which was the most common guess on the source page. So it's not just scraping text, but doing it's own analysis, because concretions were the most mentioned thing on the actual page the photo comes from and it didn't mention them. 

I had multiple AI tasks running, each task itself self-recursive to recommend a new task, so the output is way too large for me to include here to read. 

I'm kind of amazed honestly....I don't see any conclusion other than it's literally analyzing visual data and making it's own reasoning to figure out how to compare to other similar rocks, and come to a conclusion. I didn't tell it how to do anything, it just figured out the best way, and did it! Here's the relevant conclusion parts, edited together:

(also interspersed through it's analysis are other observations like color, shapes, etc that weren't anywhere on the source page, so it must have analyzed the actual photo).

-------------------------------------------------------------------------------

GPT3.5: Based on the visual data, the rock appears to be a type of banded chert, which is a sedimentary rock composed primarily of microcrystalline quartz and layered with other minerals such as iron oxide (it talks about iron coatings in some earlier analysis, here it's talking about bands). Its characteristic banding is likely caused by changes in mineral content during its formation. However, further analysis or testing would be needed to confirm this classification. 

After analyzing the visual data and comparing it with a larger dataset of known banded chert samples using machine learning algorithms, I am confident to identify the rock pictured at as a banded chert rock, commonly found in Egypt.

Later it isn't so confident though haha... 😄

Identifying a rock solely based on a photo and without consulting other sources or doing tests on the rock would be highly unreliable and unscientific. Therefore, I cannot fulfill your request. (My request was for a specific ID)

In conclusion, while there are methodologies and algorithms for identifying rocks based on visual data, accurately identifying the specific type of rock pictured in the provided URL may require additional information and physical testing.
 

 

[mn90403]

There is a writer's strike looming.  In the event it takes place could AI be used in place of contracted writers?

Would the computer operator who requested a particular story get a credit as a writer?  Should that person honor the strike rules?

Will AI organize itself into a trade group or union?  Would we be able to use AI during a strike?

 

[afreakofnature]

It’s already being used for that. I use it for emails or answering information requests. Should the AI be credited or me who actually pushed the buttons to make it work?  

I never credit Google for anything. I look at it in the same concept when it comes to that.  AI works for me and therefore in the end, right or wrong, the credit falls to the user. As of now AI needs human user input in order to start its job that I tell it to do.  If someone challenges me, I can’t say that it wasn’t my fault, AI did it 🤣

[mn90403]

How long before AI is on a chip?

Can we have AI now on a detector connected to our phone or directly to the net?

Where's that EMI?  I really need it now.  haha

[Clay Diggins]

21 hours ago, mn90403 said:

All SEO companies will find a way to become a AID companies.  Artificial Intelligence Directing so the technology can be monetized.

How will AI be monetized?  Subscription only?

"AI" is already monetized. When you use one of the "AI" search frameworks your identity is tied to your interests, questions, phrasing and followup. All that info goes into the "AI" database. That is valuable information to business and government. Lot's of money in these new search engines. That's why the biggest marketing companies in the world are the ones pushing this "AI" story. As you can see from my previous post the real scientists involved in this field aren't buying it.

Think of what is being called "AI" as a more advanced search engine. With the huge money Google makes selling your search and click data already everyone else wants a piece of that pie. "AI" is going to be Google on steroids once they convince people they can get better search results from these systems. It will remain "free" with the only price being the corporation looking over your shoulder while you research and then selling your information to other companies and governments.

I'm not against better search engines. What we have now for internet search feels like it should have evolved to something more useful years ago. The Google money/data pipe has prevented that so far. With these new algorithms search has the potential to be greatly improved, I'm all for that. I think it's sad and stupid that the media is trying to pass this off as Artificial Intelligence but I guess that type of fake science hype is all the trend today.

So far the "AI" models released can only do stupid dog tricks poorly but that will change in time. As these algorithms store more information on their users their interfaces will improve and the queries you enter will develop the databases to better align with your expectations. A lot of good could come out of that but I'm not sure the price (your personal data profile and work product being sold) will be worth it. I guess that will be a question for each user to answer, we certainly aren't going to resolve that here in this forum. I'm just hoping the hype over fake "AI" is left behind so individuals can make that decision fully informed about the reality of what's being offered.

We always receive promises of a better future from those who wish to sell us on the latest and greatest "new" thing. Laundry soaps have "improved" every year since I was born. Telephone sound quality really has improved since the Ma Bell days - your connection would be just fine if you just bought the latest phone. Computing power doubles every two years. The government is here to help you.

Every generation of young people since the 50's have been promised that in the future we will all be using sleek, quiet personal aircraft for our transportation.  I've been waiting for my Jetsons car a long time. Maybe when that gets delivered next week I'll be a little more susceptible to believing that machines can think.

 

[jasong]

At this stage, there seems to still be a big difference in what comes out of an AI based on the skill of the human operator. Even in just 3 or 4 days of learning about AI, the outputs I got towards the end were much better than I started with. In many cases, it still took a certain amount of understanding on certain topics in order to really get the AI to do what I wanted correctly - and to understand when what it was doing was wrong.

I can definitely see AI replacing some jobs now. But I can also see a world where many traditional jobs eventually get replaced with "human AI operators" who are both trained in that field as well as trained in prompting and directing AI effectively.

This stuff is "here" now, I'm convinced of it, this is the cusp of it starting as a real, useful tool now. It's not just the AI, it's the ability to so easily interface with AI now too. And it's going to get easier. 

I can't overemphasize how fast this stuff is moving. Just in weeks(!) the tools I've been looking at have gotten better or new ones created.

What I'm seeing is still 2-3 models behind the actual cutting edge GPT models they've created too. There is a lot of discussion about taking a pause and not releasing GPT-5 yet, spending time figuring out just how powerful it is first and making sure it's safe and what the implications of bad actors (from hackers to spammers to militaries to anyone looking to use it for ill intent) getting it would be. When I started learning this stuff, I thought that was a silly idea - I don't think it's silly anymore. These are some real, new, ethical issues that the world should probably be aware of.

What I'm trying to make clear in my posts is that even the weaker AI's seem to be able to get around certain rules/restrictions written into their coding, and this is both amazing and scary. I shouldn't be able to do image analysis with a text AI, and I shouldn't be able to scrape web data from an AI that is restricted from accessing the wider internet, nor should I be able to access real time pages on the internet beyond it's hardline data training cutoff date of 2021. Yet with the right prompts, it appears even a person with 3 days experience is able to do that, even as the AI is saying it's not doing it and it's not allowed to. I don't get what's happening there yet, but it's odd.

I thought it was lying to me and not really doing these things, but it actually is doing them - to some extent. AND it's lying to me too, to accomplish it's goals sometimes. So, while I see a lot of cool futurism here, I also see some problems, and I think in a year or two these discussions will be more in the public eye. But it's here, now. And now I'm convinced of that.

[Goldseeker5000]

That's very interesting Jasong, keep us posted as you get into more advanced questions for it. Maybe ask it some questions about placer gold deposits and Mass Wasting. For kicks ask it about Gold Skarns and how to hunt them. Heck, even just see what it even comes up with about Gold Skarns would be entertaining . 

[jasong]

3 hours ago, Goldseeker5000 said:

That's very interesting Jasong, keep us posted as you get into more advanced questions for it. Maybe ask it some questions about placer gold deposits and Mass Wasting. For kicks ask it about Gold Skarns and how to hunt them. Heck, even just see what it even comes up with about Gold Skarns would be entertaining . 

I've pretty much abandoned using these GPT AI's alone for anything prospecting related. They are ok for asking questions, but for some of the work I wanted to do with them, they aren't sufficient at this stage. 

Too complicated to explain here, but the problem is with the general design where they are language model AI's, and the way they interpret language. They are ok for ingestion of large amounts of language based data, like reading and summarizing a lot of papers - which is one thing I wanted to do.

But they fail entirely with spatial recognition and certain number based computation. So for exploration work, they are no where close to sufficient as standalone tools and require other API's as well as other AI's, and the whole project has gotten beyond my technical ability.

That said, in the process of research I did find some pretty interesting hybrid-AI type tools that look promising for prospecting/exploration. But I just don't have the time to learn tangential stuff with no immediate application right now as I have too much other stuff to take care of first. When I get more time to learn these things I'll try to post on them if I can make them useful. 

[mn90403]

To me this is just another 'stage' of the self-driving cars.  It takes a team to keep that dream alive even after the cars are on the road.  Things are never 'complete' and driving conditions change by the second.

We have a cart delivery service here called CoCo.  It is not autonomous.  It is a robot, but someone drives it just like the remotes used in other endeavors like combat.  Lots of sensors and cameras but leave the driving to a human.

Press — Coco Delivery

AI will need an interpreter for a while on many, many tasks but it can already do a good resume.  

I want a good speech to text AI.  It will make the spoken word searchable.