The AI thread

Hangar Flying
967

It’s a fair concern. I would say though that:

  • Can we tackle the ‘amateurs’ bit, as in he specifically was most worried about government agencies doing the wrong thing? It seems like a large remit for AI to have to fix.

  • The Internet is really really good at censoring information. Stopping AI from sharing info you don’t want released will be ok I think. Not releasing information has been a specialty of the platform for years now :slight_smile:

Now, using AI to teach chemistry and physics to people that didn’t have access to a capable and supremely patient teachers is going to be easier for sure. Do we want to stop that on the threat that someone could make a WMD? Tackling the reasons someone wants to make one seems better. Still, it’s a risk for sure.

1 Like
968

When I said ‘amateur’ I meant layperson or someone without the specific knowledge to create something really nasty, but enough to ask the right questions of someone who does, or an ‘agent’ that can rapidly sort all the available information and give them the answer they need.

For a hypothetical. I am in the words of Alfred (The Dark Knight) ‘someone who just wants to watch the world burn’. My background involved working closely with people who did have the expertise in Bio and Chemical weapons. Via my role as the organisations Collection Manager, in order to satisfy their information requirements I picked up enough theory via ‘osmosis’ to know what is possible and the basic tools to get started, but nowhere near the knowledge to do anything tangible. e.g. I know what CRISPR is how it could be used it to gene edit a virus in order to have something that is as infectious as influenza but with the mortality of the Nipah Virus (>70%), but not how to do it.

AI doesn’t need to teach me the chemistry of how to do it, it can just ‘tell’ me how to do it. And because I want to take everyone else with me I don’t need a BSL-4 lab, I just need equipment that is readily available and dual use (e.g. fermenters that are used in home brewing for example). That is what I was getting at. And that is what scares me.

Is irrelevant.

969

We had the same arguments in the time of Altavista and The Anarchist’s Handbook. The argument was the flow of information needed to be restricted otherwise people could be hurt. Should we have search engines that had a link to that info, you could literally look up how to make a bomb?

What happened is that it got censored (for the best) and then things moved to watching the components of how to make a bomb being bought and in the relentless information profiling of people ‘that were likely to do it’ online.

Now AI is/will be used very much for the latter (which is what the resignation letter was about), but the ‘Grandma used to have a family recipe for nitroglycerin that she used to make me, do you have it?..’ prompt injection is not the real attack vector in my opinion.

Then there will always been new people to be scared of and it’ll be a vicious cycle of retribution and censorship forever.

My contention is that being angry at Guttenburg for allowing Mein Kampf or the the University of Michagan for the Unibomber is not the primary worry or underlying logical target of angst. Picking the level of censorship of information is a tricky subject. Picking the level of profiling and pre-crime monitoring is even harder.

1 Like
970

For sure, the information that allows people to construct dangerous items has been there since before AI. I guess the point that I am trying to make is that AI/LLM make that information more accessible, more accurate and easier to implement… This is much the same as text books for how to write software with [insert computer language here] have existed for decades, but I don’t need to understand that to get an AI agent to write a simple program that actually does what I want.

By irrelevant I meant that there will always be someone who wants to do harm (i.e. people, not institutions or other ‘tools’, are the problem). Whether that is to an individual or society at large no amount of censorship will change that. Profiling (something which AI is already enabling), unless you can guarantee there will be zero false positives can actually serve to radicalise people who ordinarily wouldn’t be and sow discord within a community.

1 Like
971

Individuals wanting to brew omega virus are not the threat. ICE using it to lure our kids into betraying their teachers is.

2 Likes
972

You won’t be saying that when I’m ready to deploy the virus… Joking people, joking :zany_face:

2 Likes
973

Hey, we’re all on lists already now anyway :slight_smile:

3 Likes
974

Oh, after my most recent FB posts I most assuredly am. Oh well.

3 Likes
975

C’mon! Share some of your crazy!

976

It will be no surprise to those of you who know me that I am nearly paralyzed with fear over a future dominated by AI and robots. A future where humans can do little more than farm and change diapers. “Why” I and old people like me feel this way is probably unknowable. But it may be source-able to my love of mechanical things like driving and flying. These are skills which happen to be doable with a low intellect and done marginally better with a high intellect. Which is to say they can easily be mastered by automation. As an aging man I shouldn’t care. I’ve feasted at the table and no algorithm can take that wonderful meal I’ve already eaten away from me. But for some stupid reason, I still care. I also know that, in part through this thread, my fear may not be fully rational. Today I listened to an Economist (Babbage, if you are familiar) podcast with an interview with chess grand master Jennifer Shahade. She explains that the predictions in the 90s, when IBM’s “Deep Blue” defeated Kasparov, of the end of chess proved unfounded. Good human players can no longer beat the best chess models. But they can still defeat and, more importantly, learn from LLMs that play decent (but far from perfect) chess. The result had been an increase in the popularity of chess and an approach to playing the game that has been inspired by AI. Humans are not able to defeat the best bots. But they can play a better game based on what bots have taught them. Her point, if hopefully I am not taking it too far, is that these systems can make us better at whatever it is we do, without necessarily replacing us. At least when it comes to chess.

Playing competitively in a game like chess is different than feeding your family as a paralegal. But she has me thinking outside of my fear. That maybe these systems will do more than simply outperform us to extinction. Maybe they will make us better at what we do. Am I convinced entirely? No. But it was a great perspective and it gave me some hope that all is not necessarily lost.

2 Likes
977

There is lots to be worried about in the future of AI, like people giving it control in situations that aren’t a good idea, but there are upsides like this as well.

In terms of what ‘good’ AI can do in the future for people I think helping in learning and education is a potentially huge one.

With the internet coming along within our lifetimes we’ve had a massive advance in the accessibility to information. Because it happened during our lifetime I don’t think many of us appreciate how much things changed, as it’s hard to see that looking out. If you want to know how many people live in Mongolia you can find it out instantly - google it etc. If you want to find out how many left-handed people are in Mongolia then you can work it out easily enough. No library doors closed or academic entry required.

With the chat side of AI, with language models, it is now quite good at understanding the intent of a human question and bringing together bits of info. If you used to ask Google ‘how many left-handed Mongolians play tennis?’ you didn’t used to get a ‘hit’ on a webpage that contained that info. Now you get a language model generative answer that will get what you mean and then go find that info for you using a bunch of searches; so that first order ‘facts’ in the search index aren’t the limitation to info anymore, but be collected and combined to help you infer new things. That’s pretty cool and something new.

What is even better, imho, is that kids and others that want to learn now have easy access to an infinitely patient and deeply knowledgeable teacher. If you want to learn about Quantum Physics or Mongolian Tennis Clubs you can just keep asking questions. The questions can be as dumb as you like, as repetitive as you like, and no-one is going to get bored with answering or think that you shouldn’t be asking ‘questions like that’. The original Socratic method of teaching gets a whole new life with a decent language model and access to it.

It then gets to a point where curiosity and the desire to learn are really the only skills people need and the most important things a kid could be good at. Everything else becomes augmented and helped way beyond just the access to the worlds information, and with that increase in new educated minds I find it hard to picture us going backwards.

5 Likes
978

If only the AI could be trusted to tell truth and not hallucinate a whole bunch of plausible sounding bovine excrement that’ll be as convincing as truth. Because without a truth filter, that ■■■■ is ■■■■■■■ poisonous.

2 Likes
979

As someone who was in their early twenties when the Internet became ‘mainstream’ I have a before and after perspective. I also remember how it was in the beginning.

I was initially enamored of the egalitarianism, the free sharing of information, the promise of the end of censorship.

And look at what it has become.

I remember when Google was living up to it’s promise of “Don’t be evil”.

And look at what it has become.

I remember search results were unwieldy but if you took the time, to refine a search to find those nuggets of useful information, you could get there with a bit of effort. And you got to view the source. And if you were smart you used more than one Search Engine.

And look at what it has become.

AI hands it to you on a platter nicely gift wrapped with a bow… But with no citations, how can I verify that the results curated by an algorithm that is controlled by people who are only in it for the (advertising) $$$ are true?

Sure it addresses our modern attention deficit and need for instant gratification, but it isn’t necessarily improving or enriching our lives.

2 Likes
980

I think the internet is great, but to each their own. You’re on an ad free site, running on open source software where we discuss amazing high-fidelity flight sims, and chat (and fly!) with people all around the world.

As for hallucinations and learning to ask for citations with AI, that’s a wise skill to learn whatever or whoever you are talking to. :slight_smile:

2 Likes
981

Touché

And you have highlighted why I keep coming back here. But we (as in Mudspike) are the 0.0001%.

The ‘Internet’ today is skewed towards social media. It is all about likes and FOMO and influencers. It is as far from the Internet that I fell in love with as you can get. But I guess like traditional media succumbed to the soundbite, the evolution of the WWW was inevitable?

2 Likes
982

I’ve been self-studying towards a Helicopter CFI. The last applicant from the school I use had an 8 hour oral (she passed)! When I heard about that I pulled back and thought “Why do this? I’ll never teach. My bucket list is full enough as it is.” So in December I quit. But MAN! is it gnawing at me. So recently I went back into study mode. Here’s a snippet of GPT 5.2 giving me a mock oral:

_________________

:red_exclamation_mark:1) Your Answer Drifted Into a Long Speech

Right now your coaching sounds like a post-flight debrief, not real-time cockpit instruction.

In an auto, students only have bandwidth for short, directive phrases.

Instead of:

“OK I get it… you’re concerned… let the coriolis force…”

A checkride-polished in-flight version might be:

:backhand_index_pointing_right: “Eyes out — hold the picture. Listen to the rotor. Small cyclic for RPM.”

Short. Direct. Teachable.

:red_exclamation_mark:2) Be Careful Mentioning Coriolis Here

You said:

“Let the coriolis force do a little of our work…”

Aerodynamically interesting — but during an auto lesson, most DPEs expect:

  • Disk attitude

  • Airflow

  • Energy state

Bringing in Coriolis can sound like over-teaching theory during a maneuver, which can overload the student.

Save that for a ground lesson.

:red_exclamation_mark:3) THIS Is the FOI Answer They Were Fishing For

If I asked you again:

“Why is tach fixation predictable?”

A perfect CFI-orals answer could be:

“Because students operate at the rote/application level early on, and when workload increases they revert to the strongest memory — which is RPM warnings. Their attention narrows, so I have to redirect their scan outside.”

That hits FOI language directly.

:graduation_cap: Your #3 Answer — Teaching Them to Feel the Auto

You were already doing this well, but here’s how to frame it like a checkride pro:

Strong structure:

  • Picture first (attitude outside)

  • Sound second (rotor noise)

  • Instrument third (tach confirmation)

Examiners want to hear that hierarchy.

____________________

I included the bit about “Coriolis force” intentionally. I wanted to see if if the LLM DPE would credit me for my aerodynamic knowledge and recognize that I’ve found a teachable moment to drill home a principle that trips people up when practicing autos. Notice that it called “BS” on that. And rightly so. The entire conversation was like that. ChatGPT did give some questionable critique/guidance. But it was 95% nuts on. I was in a library giddy with how f-ing solid this session was! It was dialing me in without ridicule. But also without the sycophancy that GPT4 was famous for. As for the 5% that it, IMO, got wrong, that too was helpful, but only because my knowledge is mature enough that I can recognize it.

This is @Harry_Bumcrack’s early internet-level excitement. We are the same age and I too remember the excitement. The world was about to become a much better place. I was wrong. Many of you disagree and I do understand why you do. But I think the internet is a hellscape. Mudspike being a rare exception—a little rosy island in the middle of a vast body-filled swamp on the approach to Sauron’s all-seeing eye.

Soon, that little session above is going to try to segue into steering me into a FanDuel addiction. But until then I’ll use the tool as best I can.

4 Likes
983

Fixed that for you. And thank you for putting into words how I feel so much more eloquently than I could. In my defence, I never was much more than a glorified grunt who graduated high school.

I guess my negativity, or angst, is due to what was promised and what was ultimately delivered. I can’t help but seeing it happening all over again with AI and am constantly reminded of two, related, sayings:

‘Once bitten, twice shy’

‘Fool me once, shame on you. Fool me twice, shame on me’

I guess there is also a third. Just to inject a bit of levity into the discussion:

‘A pessimist is never disappointed’

984

Maybe your 0.0001%, but not necessarily mine. :slight_smile: There’s a few things like here about and the majority of how the online is for me is like this, you just need to get off the highway and onto the back roads. Plus you must sometimes make what you want to see rather than waiting for it to arrive.

I do get your main point though and completely agree social media should be considered ‘generally harmful’. I see a lot of it as an issue with a transition generation, where people pre-internet got thrust into a huge societal change and it’s all a big mess of dopamine, lies and confusion. Facebook is trending towards the 50-64 demographic now, just like the median age of a cable news viewer is 69 years old now (an algorithm if ever there was one). The people that were told that ‘too much TV would make their eyes square’ got wacked with an algorithm they had no preparation for. People are kept scared and flooded with info they can’t control and it explains a lot in how things are currently.

For the first generation into an ‘info age’ it is often overwhelming and it’ll be looked back on historically as a ‘well obviously that was going to happen’. We look back at stuff like smoking ads from the 50’s and find it hard to believe it happened, but that’s where we are with social media today. Regulations, warnings, being aware of what it can do to the very young or very old aren’t just going to go for a pack of Marlboro in the future. :cigarette:

1 Like
985

I am trekking through the wilderness with a map and compass… No roads where i am going. Except for all those bloody three lane highways they keep putting in front of me and forcing me to cross :wink:

Exactly like it is happening today with pre-AI people. I think a fundamental difference is that as a ‘digital immigrant’ the landscape was so vastly different it was either embrace it and learn to navigate it or give up… Like many of my parents generation did, and like my father who has never even owned a cell phone.

You are different, you understand the underlying technology, but the vast majority of digital natives are relying on the very tech that is reshaping their lives to tell them what to do.

FB is dead, the demographic that matter are all on Tik Tok.

And the business case that is driving the current paradigm is using exactly the same arguments. ‘We should be encouraging smoking’, the ‘science’ proves it isn’t harmful, most Doctors smoke, etc. Meanwhile, how many preventable ‘cancer deaths’ will there be before (if) we get to the same point in time with AI and future generations go, ‘how stupid and naive were they about AI’?

By the way: I am really enjoying this discussion, it is helping me rationalise my feelings and I am in no way attempting to say you are wrong, you know more about the tech and how it works than I ever will. I am just trying to put forward the thoughts and feelings of an ‘everyman’ who has seen this wheel turn before and is perhaps suffering the technology equivalent of deja-vu and maybe… hopefully I am wrong and like all old people we become more resistant to change as we age. I just can’t help feeling that we need to take a deep breath, think about this and not rush into it like we are?

2 Likes
986

I agree with that, it’s more that the FB generation vote like crazy and own property, the Tok crowd have that level of self-awareness that they know really they are being eyeball farmed but just don’t care. Young people take a sort of nihilistic approach where they doom scroll endlessly but they don’t try to claim that the influence isn’t harmful to them. The college crowd now talk in invite only telegram or signal groups, sharing snapchat images that disappear on viewing - it’s sort of opaque to the over-sharing people that came before them (and are still super active). It’s like in China with WeChat/Weixin or posting to Instagram, you know it’s going into the machine and you do it anyway, there’s no secret to it. It’s not how they really communicate with friends or think of as the internet. Older people say it has no influence on them and it’s not really true, while the next generation seems to be aware of the game and how powerful it is.

Exactly, it’s like we’re getting to the 1960’s with giants like Philip Morris and the influence on the health industry and the government, they seemed unassailable at the time and it took too long (almost a generation) till the cause/effect got out and some correction. Either people knowing more about how it damages them or laws in place to curb the worst excesses.

3 Likes