Well, nice to know that it still lies! (Sorry, "hallucinates.")

(Actually, I think it might be more accurate to describe ChatGPT's users as "hallucinating," since even people who know how it works, sometimes even its programmers, are fooled into believing that it understands things. The pathetic fallacy is...pathetic.)

_________________
Curt J. Sampson - github.com/0cjs

Me: Please write a response to this forum post:
ChatGPT4: Hello there,

Thank you for your engagement and bringing up an interesting point. You're correct in that saying ChatGPT "understands" language might be misleading. In fact, what ChatGPT does is pattern recognition on a massive scale. It doesn't "understand" language in the way humans do, but rather identifies patterns in the data it has been trained on to generate relevant responses.

The illusion of understanding is created by its ability to generate coherent and contextually appropriate responses, but this doesn't indicate any form of true comprehension or consciousness. Instead, it is a testament to the sophistication of the machine learning algorithms at work and the vast amount of data it was trained on.

As for the term "hallucination", in the context of AI, it refers to the model generating outputs that seem plausible but are not grounded in reality. This can occur due to the inherent randomness in the model's design or the mismatch between the training data and the real world.

Your comment about users or programmers being "fooled" into believing that the AI understands things is a good reminder for us all. It's crucial to remember that even the most advanced AI, like ChatGPT, is just a tool that operates based on its programming and training data. It doesn't possess consciousness, beliefs, feelings, or an understanding of the world in the way humans do.

We appreciate your input and critical thinking in understanding the capabilities and limitations of AI technology.

Best,
ChatGPT

_________________
http://WilsonMinesCo.com/ lots of 6502 resources
The "second front page" is http://wilsonminesco.com/links.html .
What's an additional VIA among friends, anyhow?

_________________
x86?  We ain't got no x86.  We don't NEED no stinking x86!

This stuff isn't artificial intelligence, it's just rule-based control systems that are very predictable. It's just that you have to understand what the systems do which is complex. But it's a general design rule of aircraft that there's always a way for the pilot to override automation, except in cases where a pilot really can't do the job (such as an aircraft with negative stability, which would not be anything used in commercial air transport) or, these days, when the commanded changes would exceed aircraft frame limits (i.e., it won't let you do things that make the wings fall off, though these protections may be removed if enough flight systems fail that it can no longer properly detect such commanded actions).

I can't speak to the specific situations you mention without references, but I suspect if you read the NTSB (or similar authority) reports, you'll find that your view of what actually happened will drastically change from, "the aircraft forced the pilots to crash."

Remember, autopilot and suchlike isn't replacing pilots. It's just another control method to let a pilot fly an aircraft with less work. But of course you need to do more work in training to understand the automation.

_________________
Curt J. Sampson - github.com/0cjs

_________________
x86?  We ain't got no x86.  We don't NEED no stinking x86!

Well, I'd certainly disagree with that. While ChatGPT isn't, to my mind, AI (I, rightly or wrongly, associate "intelligence" with the ability to reason), SHRDLU to my mind clearly exhibits the basic characteristics of intelligence.


Really? I'm not an Airbus pilot, but my impression of Normal Law is that such behaviour would be perfectly predictable. Maybe not what you want in that particular situation, but predictable nonetheless.


Except that autopilot has always been a computer, though not always a digital one.


Well, you may consider it to be "too often," and many would agree, but are you looking at the overall incident and accident rate, rather than cherry-picking anecodotes? The stats tell us that commercial aviation is safer than it's ever been. We can't prevent every problem, but trading a few accidents that did happen for hundreds of accidents that didn't, but would have under earlier control schemes, doesn't seem to me such a bad thing.


This is true; it's a known problem that we can see we need to take steps to address. But again, "mistakes are being made" is the wrong thing to look at. The right thing to look at is, "how many mistakes are injuring and killing people now, and how many were injuring and killing people under the old scheme."


Yes, very true. On the other hand, people have already been killed by strong airmen who made mistakes that these days automation catches and corrects.

Humans make errors, and "be a better pilot!" is not going to fix them all. All the evidence points to the current trade-off we're making being better than the previous one.

_________________
Curt J. Sampson - github.com/0cjs

Of course, this is the same argument that fully controlled cars will replace manual drivers because *most of the time and for most people* they do a better job of it. Right up to the time they don't see a heavy goods vehicle across the road, or mistake an exit barrier for a road, or can't see a bicycle...

Neil

Right, there will still be accidents, just different accidents than would have occurred without the automation. Even so, if and when we get to the point that autonomous vehicles are proven to be significantly safer on average, it will be hard to argue against wide adoption.

_________________
x86?  We ain't got no x86.  We don't NEED no stinking x86!

_________________
x86?  We ain't got no x86.  We don't NEED no stinking x86!

We have gone utterly off-topic. There's usually a cultural difference between a specialist forum and general purpose social media.

_________________
x86?  We ain't got no x86.  We don't NEED no stinking x86!

Well, some people will find that a good argument against automation, but they're terrible engineers and should not be in charge of this stuff.

Basing one's approach to safety on emotionally resonant anecdotes literally kills people. Germany has done this over the past decade since Fukushima when they decided to shut down all their nuclear power plants. A good chunk of that power generation capability was replaced by coal which is far more polluting, directly kills far more people per megawatt generated even before you get to things like earlier deaths from pollution, and, ironically, releases more radiation into the environment per megawatt generated even counting radiation released by nuclear plant accidents. More people will die because of this than if they'd continued their use of nuclear power. (But at least they'll have died in the "right way." Sigh.)

That said, we're a long way away from reliable fully autonomous cars, much less aircraft. Arguing against that one is easy because the technology's not there yet. (Consider that we haven't even really started in on fully automated aircraft taxi yet, a much easier job than either, and one that would save airlines a considerable amount of money.)


Well, your limb broke off long ago, because we've been replacing human judgement with automation for a long, long time now. I use it every day: I, as with literally tens of millions of people around me every day, rely on my rice cooker to cook my rice rather than monitoring it myself, and it works much better than if I did it. We've been doing that since the 1950s, and it had virtually completely replaced human judgement about temperature settings by the '80s.

In the realm of safety, automation is widely accepted too. There are a huge number of examples of things from anti-lock brakes to GFCI where we let automation completely take over because it indisputably keeps things a whole lot safer than letting humans do the job does.

_________________
Curt J. Sampson - github.com/0cjs