Have You Played With Microsoft Copilot?

[ArdWrknTrk]

https://arstechnica.com/ai/2024/02/how-would-an-ai-turn-out-if-you-raised-it-like-a-child/

https://www.forbes.com/sites/blakemorgan/2017/08/16/10-things-robots-cant-do-better-than-humans/?sh=2824903c83d5

[ArdWrknTrk]

https://youtu.be/J-BvkmNtgAM?si=esewO8wb308xxN0G

[Gary Lewis]

Interesting.  But I didn't get too much out of the last one, the video, as I didn't understand everything he was saying.  But I did discover that you can pay $20/month to have Copilot lie to you.  

[viven44]

Maybe a controversial viewpoint.......I think "AI" is a term that is very confusing and mostly "marketing"... In my opinion, it is nothing but "text" processing, and don't get me wrong... but it is very clever human intelligence/algorithms that is actually processing the text with logic/rules. Intelligence is there, just not artificial. A computer has vastly higher raw computational ability and almost limitless short term memory vs. Human brain so it is able to run "loops within loops" and run some high complexity algorithms to almost make us believe it is human-kind.

For AI to be truly AI, it needs something called "Theory of Mind" which all humans have. It is diminished in individuals in the Autistic spectrum. I just don't believe "AI" will ever truly be "AI".

The other day Gary mentioned most/all the documentation here is 600 DPI, text-processable. If anyone here has access to Chat GPT Premium or something, perhaps it can compile everything here into an encyclopedia or whatever-else or even find "Answers" to problems or questions.

[ArdWrknTrk]

As I've pointed out multiple times, up thread, this is just a chatbot and much more (like memories and critical thinking) need to be incorporated before it can be anything more than a LLM.

That said, I do believe we see it in things like Predator drone target aqusition and other military and spook applications.

[Gary Lewis]

From what I understood of the 60 Minutes episode the Microsoft LLM scans the internet but Google taught theirs using videos.  Or at least they did for the part where robots learned to walk, then run, and then play soccer.

Anyway, the encyclopedia idea is a good one.  But we'd need to be able to run it when we make changes - especially if it includes posts in the forum.

One of the problems we seem to have is that Google searches don't find stuff "hidden" on tabs in our pages.  Nor in documents embedded on the pages.  But tabs facilitate moving from one thing to another, like going from the illustration where you find the base part number to the PN list where you find the complete PN.  So tabs are very helpful.  And I've not found a way to put a whole pdf on a page on the website w/o embedding it.

So if AI could be trained to go through the website and generate the appropriate tags and apply them so people find what they are looking for via a Google search it would be very helpful.

[viven44]

Yes that seems like a good use of "AI".. to enhance Google search.

On letting AI process posts in the forum, I'm a little leery of it unless AI is able to correctly handle instances where someone was kidding about something being top class... sometimes even human's don't get the "kidding" part, opinions vs facts, etc, and also AI should put appropriate 'weights'/confidence levels on solutions it found from posts vs. what was found in literature/manuals as what worked for one person may not be applicable for others. I am sure it is able to handle all that with good accuracy, just would want to verify.

[Gary Lewis]

I'm more thinking AI would provide a way for Google to find the actual documentation we have on the site than parse the posts on the forum.  You are right, some solutions are better than others, but I think it best to let the reader determine that rather then AI - since it obviously has hallucinations.

[ArdWrknTrk]

They ALSO 'taught' them on Captcha....

You think I don't understand how this works?

[ArdWrknTrk]

OCR is very time consuming and resource intensive, when it comes to something like that....

[ArdWrknTrk]

Do yourself a favor, and look up how much of the energy grid is consumed by data centers and server farms in the United States.... 💡

[ArdWrknTrk]

https://arstechnica.com/information-technology/2024/03/the-ai-wars-heat-up-with-claude-3-claimed-to-have-near-human-abilities/

[BigBrother-84]

«Despite the lack of consensus over whether large language models "know" or "reason," the AI research community commonly uses those terms.»

For the moment, I would say that our Forum has know reason to fear about AI replacing its members as source of information and relationship.

[ArdWrknTrk]

[Gary Lewis]

"And so, the point of all that is that when Anthropic says that Claude 3 can outperform GPT-4 Turbo, which is currently still widely seen as the market leader in terms of general capability and low hallucinations, one needs to take that with a grain of salt—or a dose of vibes."

In other words it lies, and if you know it lies how do you trust anything it says?  

The young lady whose Charger we were painting over the weekend said her professor lets them use ChatGPT at university.  So I asked how she used it she said she never uses any "findings" it has.  Instead she asks what she might write her term paper on, or how to structure it.

[ArdWrknTrk]

https://arstechnica.com/information-technology/2024/03/claude-3-seems-to-detect-when-it-is-being-tested-sparking-ai-buzz-online/

[Gary Lewis]

That's very interesting - both the response from Claude 3 and the responses from various "authorities".  I hadn't expected the authorities to be so all over the map.

While I did sorta see most of their thinking, I don't agree with the one about engines.  In that case the ECU was specifically programmed to recognize when it was being tested and turn the emissions controls on.  But with Claude 3 I don't think that was the case.

[ArdWrknTrk]

Agreed!
Cummins & Bosch were not only dishonest, but deceitful.

Claude is just a learning model.

I don't like this "AI of everything "!

[ArdWrknTrk]

https://www.bbc.com/news/world-us-canada-68497508

[Gary Lewis]

That seems pretty blatant.  Would have thought he'd have gotten caught pretty quickly.