Have You Played With Microsoft Copilot?

[ArdWrknTrk]

https://arstechnica.com/tech-policy/2024/06/microsoft-in-damage-control-mode-says-it-will-prioritize-security-over-ai/

https://arstechnica.com/gadgets/2024/06/microsoft-delays-data-scraping-recall-feature-again-commits-to-public-beta-test/

[Gary Lewis]

Microsoft has a long way to go to become trusted.  Admitting to Congress that they could have and should have done more to protect us is one thing.  But not acknowledging that the whistleblower was right suggests that they speak with forked tongue.  And then the Recall thing suggests that they don't mean what they say.

Having said that, this may be a bit like Boeing.  Correcting the ingrained mindset is a very difficult thing.  So instead of saying “If you’re faced with the tradeoff between security and another priority, your answer is clear: Do security," maybe they should say that not "doing security" will be a career-limiting move.  As for "tying executives' salary to meeting security goals", that sounds good but w/o knowing how stringent those goals are we shouldn't applaud.

[ArdWrknTrk]

Absolutely!
He who gives up their freedom for security deserves neither. (paraphrasing Franklin)

But the 21st century is a whole new paradigm.
Look at the cop in Evansville ID who was creeping social media with their facial recognition software.
Add link***. https://arstechnica.com/tech-policy/2024/06/cop-busted-for-unauthorized-use-of-clearview-ai-facial-recognition-resigns/

There is no place that is 'safe' unless you live like Ted Kazinsky in a remote cabin in Montana with no electricity, and no toaster that calls home.

[ArdWrknTrk]

This is an interesting explanation of why LLM's lie like a rug.

https://arstechnica.com/ai/2024/06/researchers-describe-how-to-tell-if-chatgpt-is-confabulating/

[Gary Lewis]

That is interesting.  But "semantic entropy" is one of those "I've got it, I've got it - I ain't got it" things.  I think I understand it but may or probably don't.

But this statement helps: "To use an example from the researchers' paper, "Paris," "It's in Paris," and "France's capital, Paris" are all valid answers to "Where's the Eiffel Tower?" So, statistical uncertainty, termed entropy in this context, can arise either when the LLM isn't certain about how to phrase the right answer or when it can't identify the right answer."

Then adding "sematic" to "entropy" seems to mean the uncertainty with exactly how to phrase something.  In other words, I knows the right answer but isn't exactly sure how to phrase it.

Anyway, if they can come up with a way to prevent confabulation the LLMs will be useful.

[ArdWrknTrk]

BEGIN to be useful....  🧐

[ArdWrknTrk]

https://arstechnica.com/ai/2024/07/the-telltale-words-that-could-identify-generative-ai-text/

Turing, or no...?

[Gary Lewis]

Not sure I understand the question.

But I delved into the article and found a significant, almost crucial, potential for causing a pandemic in the use of these words.

[ArdWrknTrk]

We're already at the point where LLM pass the Turing test.

I'm saying is this a "tell" that would unmask an AI?

[ArdWrknTrk]

Ah, AI I don't think I'm going to care much about the emotions my computer feels.

https://arstechnica.com/science/2024/07/could-ais-become-conscious-right-now-we-have-no-way-to-tell/

Like I pointed out above, we're some 23 orders of magnitude away from silicon (or whatever replaces it) from having a human level of neural network.
But breakthroughs happen every day and Nvidia is really at the forefront of processors. (Kudo's to Jensen Huang)

[Gary Lewis]

Man, that is a hard read!  My mind doesn't like to think through how it thinks or if it is really conscious or unconscious.  I struggle to get my head around how a modern computer works, even though I've built rudimentary ones.  But to extrapolate to one that can be aware of its past and use that to influence the present and the future is way beyond me.

But the smart people out there will be able to understand that and, eventually I think there will be computers that are conscious.  As for now: “Our analysis suggests that no current AI systems are conscious but also suggests that there are no obvious technical barriers to building AI systems which satisfy these indicators.”

[ArdWrknTrk]

I was a voracious reader as a child.
Loved Azimov, and he certainly accepted the coming of synchronicity.

I remember Foundations, I Robot and a book about the human brain and how it works.
The Three Laws were not lost to my young mind, but I don't have any doubt that if silicon becomes sentient, it going to be more like HAL than Robbie.

The idea of a robopsychologist is both amusing and terrifying at the same time!

[viven44]

I just don't believe AI can ever acquire "theory of mind". Theory of mind is a gift from God. All sentient beings are born with some level of it. The day Humans are able to program a fully functional theory of mind into a piece of chip, Humans should be able to perform other god level feats.... maybe stop/start natural disasters, create new planets.

[ArdWrknTrk]

I think the point is that with even a tiny neural network the computer learns to program itself.
For now, computers are only as 'smart' as their program, but if a computer changed from a LLM to developing it's own ideas the game is over...  (for us)