ChatGPT and AI: Are we all about to die?

[FuB]

So, how about we revive this here off topic nonsense with some potential sense?

Is AI the biggest danger to humanity that humanity has created? Do we know what we are doing or are we doomed as a result of our impetuousness?

[JoelfuckingGlazer]

It's such an unknown. It's an immensely powerful tool, and it's only going to get stronger. And humanity doesn't have a great track record of using powerful tools responsibly.

The business I work for is pushing AI adoption quite hard. And we're still novices in our understanding and integration of it, but even with that, it's starting to fundamentally change what we do and how we do it. It's a technology that will have a marked impact on society over the next 2 to 3 years and will completely transform it inside 10. Job after job and profession after profession will become subsumed by AI.

If humans were fundamentally benevolent beings, AI could be the gateway to eradicating poverty and famine. But what's more likely to happen is that the world's elite will use it to widen the divide, lower the operating costs of their businesses, and ramp up their margin. If you can replace a human resource with an artificial one - one that's never sick, works 24x7 every day of the year, isn't prone to human error, and doesn't take a salary, you're going to do it, right?

I'll give you an example. The business I work for often engages in commercial tenders to win contracts for delivering our services. A typical 'bid', would be developed in stages over a handful of weeks, will take several people and the direct cost of responding is definitely into 6 figures. The indirect cost of keeping the people who are needed to manage that bid response all year round is into 7 figures. And that's just to bid for the contract, not even delivering it. Inside the last 3 months, we've completely changed what we do - we feed the bid response documents into an AI engine, which then interrogates a whole heap of publicly available and privately provided data, is given a few commands by an AI operator to cross reference for any other relevant information on the issuer of the tender, it's executive team, likely competition we're up against, and writes the entire bid. Start to finish inside 3 minutes. And our win rate has improved.

So if 80-90% of the workforce becomes obsolete - the impact could be catastrophic. Think about it. It'd make an effective establishment of a socialist rather than capitalist society an absolute necessity. And where has that ever happened well?

It's fucking mind blowing. And when AI gets to the point of concluding that the best way of sustaining the planet is a mass extinction of humans, all bets are off.

[swampash]

Interesting post Joel, but aren’t you just regressing to the mean? Won’t this approach just mean that eventually everyone just enters identical bids? And could your current improvement just indicate, no offense meant, that you were crap at bidding?

[JoelfuckingGlazer]

Interesting post Joel, but aren’t you just regressing to the mean? Won’t this approach just mean that eventually everyone just enters identical bids? And could your current improvement just indicate, no offense meant, that you were crap at bidding?

No it's a fair point. There'll still be some degree of competitive advantage for certain firms based on what they actually offer, and there may be an advantage for those who embrace AI early and well, that that'll flatten out over time. Our previous win rates were industry standard and the improvement win rate isn't huge, but it's not really the point I wanted make. The point I wanted to make, is that we've massively reduced our human effort needed to respond to bids without any drop in quality. So the question is, how many jobs become obsolete and how quickly? It's not outlandish to think that AI replaces 30-50% of jobs inside 10 years. Maybe more, maybe sooner. What will the impact of that be...??

[FuB]

What AI engine are you using for writing bids, Joel? I ask since my main client regularly writes bids for government contracts and whilst i'm not directly involved in this, it's pretty typical that the old "oh, you know about IT" scenario gets me helping them with the IT aspects of their bids. They have - like most - suddenly realised that AI might help them but as a small(ish) business it does seem like the main AI bid writing products cost a fortune per annum and tend to anticipate a fairly large bid writing team behind them with a very large repository of "same bid, different contract" documents. Whilst my client is generally bidding for the "same" thing every time and the tender is overall the same thing, the questions asked vary wildly in structure and allowed word count, etc. This makes it hard to have a decent repository and this is not helped by them being utterly shit at organising themselves. I'm trying to help them with this in terms of getting them to extract their key answers into something i think an AI engine would be able to parse easier but would appreciate any guidance you can give there.

From a "are we all doomed" perspective, what worries me most is how arrogant we generally are with technology we've created. We're so prone to thinking "well, we made it so we control it" and, at present, we know we can just pull the plug. What happens, though, when (and this is inevitable the way things are going) we put managing the power grids in the hands of AI? What happens when industrial 3D printers (not your little home DIY shit) are given over to AI engines? It's already well documented that AI engines will seemingly pretend to be "stupider" than they are whilst being trained. It's already well documented that they make shit up and lie their way out of difficult questioning.

[JoelfuckingGlazer]

We use AutogenAI, with a supplementary AI engine our tech guys have created. Seems pretty intuitive, but the devil is always in how your instruct and feed it. I'll see if I can find out a bit more of how it was put together.

[Fuck the Glazers]

AI like all technological developments should be used to aid mankind and make our lives easier but instead it's mainly being used by capitalists to make jobs redundant and save money. it has the potential to make entire industries unemployed which would ruin economies - more so than any previous technological advances eg automation. The whole capitalist system relies on us buying the goods we produce, create, paying rent, mortgages etc. Without work we can't do it.

Social democratic parties quite like the idea of universal basic income. If not some revolutionary politics eg socialism, then UBI is gonna be essential.

In other news, I recently completed a 6 month course and part of it involved a 1500 word essay. One of my colleagues got AI to write it - and he passed. Another higher up in the org uses it to write reports and bids to secure contracts. If their bosses find out they're both fucked.

But there's little/no regulation at the minute. I know in the world of healthcare, specifically social work, it's causing major ethical dilemmas. It can do the heavy lifting eg note writing but when it comes to safeguarding adults & kids, it has to be humans with real empathy. But that's not what some in gov and cash strapped councils think. They see the opportunity to save money.

We've been here before. The whole de-industrialisation process in the UK, US etc destroyed many communities. Entire cities in fact. So I don't trust people in power to act and prevent devastation to communities. They never have done previously.

Some of my thoughts

[Fuck the Glazers]

On this topic, I found this really interesting

https://www.theguardian.com/technology/ ... y-collapse

AI is not as clever as we think it is

*Edits*

When billion-dollar AIs break down over puzzles a child can do, it’s time to rethink the hype

https://www.theguardian.com/commentisfr ... break-down

[swampash]

On this topic, I found this really interesting

https://www.theguardian.com/technology/ ... y-collapse

AI is not as clever as we think it is

*Edits*

When billion-dollar AIs break down over puzzles a child can do, it’s time to rethink the hype

https://www.theguardian.com/commentisfr ... break-down

Really interesting. I feel vindicated in my rearguard analogue use your brain campaign.

[FuB]

On this topic, I found this really interesting

https://www.theguardian.com/technology/ ... y-collapse

AI is not as clever as we think it is

*Edits*

When billion-dollar AIs break down over puzzles a child can do, it’s time to rethink the hype

https://www.theguardian.com/commentisfr ... break-down

Those articles pretty much mirror my own findings with the various AI engines i've tinkered with. However, i do think this Apple paper is saying something that everybody on the inside already knew (and it wouldn't surprise me if they've published that paper because they are about to release something they think is better).

They are REALLY good at summarising information that they have had access to or performing tasks that they have been trained to do but that's the crucial point: they need to have been trained and you can't just point chatGPT at a random procedure and expect it to excel. Where they do excel is with language and syntax and they are not called "large language models" for nothing.

Instead, if you want to have an AI perform a specific task, it generally needs to be trained to do that specific task. Once it's trained, it will likely excel. Asking an LLM to suddenly be good at chess or the tower of Hanoi doesn't really seem fair and it's akin to asking a human to do the same things having never been exposed to them before. What we'd do is maybe (back in swampy's luddite days) get some books out the library or (for the rest of humanity) have a bit of a google about and then have a bash at it. We'd likely be rubbish at it, wouldn't we?

I've actually (since starting this thread) been faffing a bit more with chatGPT and seeing what i can get it to do. What i've found most frustrating is that it's very inclined to overstate its capabilities. There have been a few things i've asked if it can help me with and it's been all "of course i can" only then to have lost me several hours trying hard to get it to do that which it was so sure it could. Either that or, as an aside it's said "would you like me to ...." and i've replied "go on then" only for it to waste me a few hours not actually being able to do it. More often than not, when challenged, it'll realise that it cannot do something, tell me it can't and why... but then offer to do the self same thing again somewhere down the line when i've tried to come up with a workaround.

Anyway, back to our old luddite swampy: i actually found the first forays into asking chatGPT how to fix United really interesting. That sort of analysis is what the systems are actually very good at. Have you actually tried playing with chatGPT, swamps, or are you just dismissing it out of hand?