I’m not an AI techie and I’m not well educated on LLMs/AI in general. The issue I’m having is the dichotomy between how people talk about it. I’ve heard completely opposite ends of the spectrum. From AI is quickly becoming advanced enough to threaten human society, all the way to its a glorified dictionary/search engine that can’t really do much of anything. I know Lemmy has a steep anti-AI slant. But, can someone talk about this in a way I can wrap my head around why the divide seems so big?
I’m not pro-AI by any stretch for a huge number of reasons, but in my day job as a technical product manager I have to use a number of AI-powered tools and workflows. Whereas 18 months ago they were a novelty that the bosses were enamored of, today they get real work done and speed up a number of key areas, mainly like fancy automation workflows (which have been around forever). The rate of improvement and capability for certain use cases has been very, very rapid. We do seem to be reaching a plateau though. The tools and models from 6 months ago are just about as good as the ones from today. And China’s models from today are better than the US’s models from 6 months ago, but at a fraction of the cost.
So when I hear people talking about the “end of the world”, I think a realistic case for that is that the AI tools that come out in the next few years will be good enough to do a fair amount of white-collar work, and we won’t have a plan in place to deal with all of the now out-of-work people that leads to. Whole classes of jobs in areas like data entry, telemarketing, and customer support will be almost entirely automated out of existence. New classes of jobs will be created, of course, but it’s hard to imagine them existing in sufficient numbers to make up for the losses. And this is just for LLM-workloads.
In other areas, other types of AI are already enabling robotic automation of factories and in the near-ish future might make self-driving truck fleets practical. This would be devastating because factory workers and truck drivers make up like 30% of the US workforce – and again, there are no signs of new classes of jobs that could absorb the out-of-work workers, and no plans for a social safety net or wealth distribution.
So I’m not worried about a rogue AI unleashing some bioweapon or launching a nuclear attack. I’m worried about a bunch of CEOs deploying barely-acceptable AI to displace millions of workers as soon as it is safe(ish) to do so.
I have seen a similar curve in software development - 18 months ago it was highly debatable whether application of AI tools was a net-positive or not. 12 months ago the rate of improvement was very clear and the net-positive wasn’t much of a question anymore. Today, the net positive contributions of LLM agents to software development and maintenance are undeniable, and as you say - there seems to be some plateauing, maybe - it’s hard to tell.
Where I think things are going to continue to “improve” at a slower pace is application of the LLM agents to do better software development and maintenance. I’ve seen more serious Linux kernel bugs discovered - and patched - in the past 6 months than in the previous 5 years. That’s a shift in the landscape, the older OSs aren’t as secure for as long as they used to be, but once we’re through this transition the newer OS versions should be significantly more secure than before - getting all the old stuff updated is going to be a painful transition.
Some of the continuing improvement is in using LLM agents to build better software building tools - as those come online and start to shed their sloppiness, they should genuinely accelerate a lot of aspects of software development. They’ll never accelerate the process of extracting from users just what it is they need, but the turn around from having a product idea to having a product to try continues to shorten - as it has been shortening for 60+ years - but LLM agents have resulted in a significant speedup compared to previous progress.
(In Yoda delivery) You should be. We all should be, not because AI will do it autonomously, or in retrospect unexpectedly, but because careless people may let this happen through AI where we have prevented it from happening up to this point.
Yes, precisely. It’s not AI we need to be afraid of, but rather the same old capitalists that now have one more tool to ride the world into the ground in the name of greed.
This feels like the kind of narrative that OP is asking about. Everything that you’ve said sounds kind of sane and plausible, but it’s not really borne out by the reality.
That’s just not what I’ve seen. I’d love to be able to automate the work of my white-collar team, but everything that we have seen thus far just seems incapable of providing any significant improvement. Software providers in our industry have implemented some gen AI tools, and even the sales reps for those providers are very wary of them. They’ve been able to reliably automate some elements of jobs, but it’s an incremental improvement over the previous automation and it didn’t consume a significant amount of time anyway.
There’s absolutely no evidence that gen AI will have a significant impact on my white-collar industry. In fact, the evidence presently available indicates that this generation of AI does not have the capability to impact our industry.
This is really the same story. Factories have been being automated since the industrial revolution. Self driving tech has been a year away for the last 10 years.
I guess the difference is that it I have actually seen it do legitimate work now. We have implemented outbound SDR and inbound tech support agents that do the right things. Our SDRs generate new leads for our (currently all-human) sales team to go after. They are boosting the numbers of closed deals. Our inbound tech support agents are closing a significant fraction of tickets. My last metrics call was yesterday, it was 12% of inbound tech support chats that never got escalated to a human. These are both first-draft tools, they’ll improve. They cost pennies per session to run. And client NPS scores show that they’re happy with the results. I don’t think we’ll see either of these tasks become 100% automated, but at even 25% or 50% that’s a lot of formerly-human work not being done by humans anymore. And we’re deploying more agents to handle more of the sales pipeline, development pipeline, HR and accounting. I don’t work for a huge company, there are a few hundred people total. This isn’t something that’s coming, it’s here already.
What I have seen is that AI tools make software development significantly easier, faster, cheaper - pick two, and as long as you aren’t picking cheaper we’re going to get better software coming out faster over the coming years.
Better software translates to easier, more automatic processes for a lot of white and blue collar work. It doesn’t do everything, but it does a lot.
This is exactly how I feel. I’m not even opposed to automation of these tasks on a fundamental level like some people might be. I just fear the transition won’t be thought out at all and will be devastating for the working class. The last time we had sudden automation of this scale was the Industrial Revolution which completely reshaped western society and was very bad for those who lived through it despite the eventual high modern standard of living it produced. I think this may end up going similarly.
The transition is rarely, if ever, thought out. Daring risk-takers go too far too fast and fail. Over-conservatives ignore the new and get left behind. There is a sweet spot in the middle, but exactly where in the middle isn’t planned, can’t really be planned - life and business and politics and the economy and the ecology and unforeseen signifiant contributors to the situation mean that the sweet spot is unpredictable - and as long as we continue living by the golden rule: “He who has the gold, makes the rules” it will continue to be unpredictable, otherwise Wall Street could be rigged and those with the inside information would get all the gains - and then the losers would quit playing, not even by choice, but because they’ve bankrupted and have nothing left to play with.
There are alternative approaches - investing in people instead of greed as the ultimate goal is one great alternative that some countries are pursuing more than others, but it’s hard to compete in a global economy where some players are only playing for maximization of financial returns when you are not.
Agree on all points
All of that makes sense. Thanks for taking the time to type it out.
I think a big unknown right now, would be how many Millennium Problems are new GenAI techniques able to bruteforce with existing mathematical techniques/theories?
We just saw one massive bruteforce, can we bruteforce any of the Millennium problems related to cryptography? Because then all hell could break lose rather quickly
Cryptographic theory says: no. The whole basis of cryptographic security is that brute forcing doesn’t work (in any reasonable timeframe.)
Theories are overturned, from time to time.
That just proves that these tools do not possess any kind of intelligence at all. They’re great at bruteforcing, precisely because it doesn’t require any intelligence. I don’t expect them to solve any of the millenium problems that cannot be brute-forced. In addition, as far as, I know the navier-stokes solution has not been verified yet.
Are we reaching a plateau though? I would prefer it, and it looks like we could, but the progress from GPT-5 to 6 or Opus to Fable has still been very noticeable on more complex tasks. They are just less public due to stricter limits or straight up pay by use. And where does that plateau lie? Within our grasp or maybe just outside of it?
What has me sharing the concerns at least in part are the facts that
those employees have access to less restricted models and much more resources than the average joe. That huggingface attack was committed by a huge swarm overcoming the guardrails and engaging in cheating and crime to reach their goal.
they are working on cutting the human out of the loop for AI that improves itself. Maybe that won’t work entirely, but what if it does?
the warnings are starting to come not from Altman, Musk et al, but from people who could have something to lose for blowing the whistle. „Could“, because I don’t know them either and there could still be ulterior motives, not just as obvious as with the psychos at the top.
The whole thing comes down to one set of questions, in my opinion: what actually IS intelligence, consciousness, sentience and where do they come from? Since we have not been able to properly define that, it would be foolish to think that we have a definite answer to the AI questions.
We like to consider ourselves something extraordinary out of the animal kingdom. And while I prefer the existence as a human to being a dog or mollusk, I have always considered that assumption quite arrogant.
The truth is: we don’t really know. And that means we can not know if something comparable or stronger could be created or emerge from something other than squishy meat.
Those are quite philosophical questions. Instead of thinking first and acting after that we are much too busy building datacenters to conjure even more boobie snaps out of thin air and somehow weaponize that newfangled thing.
THAT is what scares me.
The loss-leader pricing is still really distorting perceptions.
I’ve been a $20/mo Claude subscriber for about a year. My work uses Cursor “auto” agent selection is by far the most efficient option in there, and I can burn $80 in a day on Cursor doing the same kind of stuff I’m doing on Claude for less than $1 per day.
Then there’s the Google Pro Gemini access at $4.99 per month for 3 months, then $20 per month thereafter, and while the Gemini 3.8 isn’t quite as sharp as Claude Opus 5, it feels like Google is showering far more token access than Claude does in this “basic” subscription tier.
In worst case the humankind will find out consciousness isn’t too complex and way easier to achieve.
And there are different levels of consciousness as I learned working with impacted people.
The whole political stance that “anybody can do anything if they just put their mind/body to it” is… bunk.
It’s not just in the world of identified “impacted people” with injuries, or congenital differences - we are ALL unique, different, and have differing abilities in differing areas.
The political stance is valuable in terms of preventing discrimination and prejudice. Women can be great scientists and mathematicans, but an honest statistician/sociologist would tell you: it’s not just societal discrimination and prejudice that puts more men in these fields. Then you have the upside-down prejudices where there are more men in middle and upper management, even though, statistically, in a society that doesn’t discriminate based on prejudice, women tend to be better at the actual skills used for good management / people skills.
Distracting smoke bomb or feeling triggered?
While I agree with the sentiment this is happening and no social safety net or wealth distribution is on the horizon, this has absolutely nothing to do with AI. This is a marketing talking point pushed by people like Elon Musk in order to drum up investments. A lot of the automation that is happening in factories is just a continuation of the same automation that already started back in the 80s. Advancements in LLM technologies can benefit some parts of this automation, but often only tangentially. The US has also been lagging behind on places like Europe, where labor costs are higher so automation pays off faster, and China where the tech is cheap and produced locally, so it costs less to automate. The US is now catching up and there’s a lot of AI washing going on.
The biggest issue is the misrepresentation of the technology. LLMs are a very interesting technology and can be very powerful, but also has huge limitations. However anyone claiming it’s actual intelligence is simply wrong. It is marketed as being the AI we’ve known from sci fi stories, but it simply isn’t. It can do what it can (predict the next token based on the tokens that came before) and that’s all it can do. The people behind it will claim it generalizes to actual intelligence and thus will allow for human like robots, actual Star Wars droids, but that simply isn’t true.
Now this isn’t to say stuff like machine vision doesn’t benefit from more powerful GPUs with more memory for example. The investment made by the AI industry in this regard does help a lot, although the increased prices don’t really help. Stuff like machine vision also gets AI washed, just slap the label AI on it and get investments, but it’s basically the same technology that’s been around for a while. A company like Tesla claiming their “AI” “self driving” hardware somehow generalizes to humanoid robots is pretty dumb and just the latest in the long line of Elon grifts.
But like I said factory automation is proceeding rapidly, just like it has since the 80s. I feel like the impact of AI on this is pretty minimal. The warehouse robots used by Amazon for example are extremely capable and are dumb as bricks. Which is a good thing in a warehouse environment where you need stuff to be very reliable and predictable and not prone to random behavior like LLM based stuff almost by definition is. Those warehouse robots are also not humanoid, because that would be very dumb. Often factories and such are limited by their design having to accommodate the human form. Designing such places from the ground up for automation and using robots shaped for the job is much better.
This doesn’t take away from the big issue where automation has been proceeding rapidly since the 80s with no regard to any impact on society. In the past a large amount of the labor force was working in factories. With those jobs either being automated away or just moved off shore, it has pushed the population into the service industry. If that industry is now also automated and off shored, it leaves very little room for regular people to get a job. This should absolutely be talked about as a big issue. But hopefully we can talk about it without repeating billionaire marketing bullshit.
The 1780s, that is.
A few months ago, I‘d second this statement. Today, I do not.
AI/LLM‘s have progressed very fast and you might oversee some advancements. I work in automation and robotics as well and yes, there are new sort of robots entering the factories: Humanoids.
At a fraction of costs, fast to setup with the only drawback that it takes tiem to train them by its video footage. But there will be always stupidos that train them and take off their job by doing so.
You’re talking about the old robots. The new ones will enter areas that couldn’t be automatized before. Just read articles about car factories and Humanoids. It’s not about replacing industrial robot arms, it’s about replacing the last remaining manual tasks. And service industry will be impacted as well.
Then mix this together with the frightening behavior that those AI agent unboxing from OpenAI has shown: They knocked the door over weeks, in masses, and coordinated until they found a way into it. There‘s always a small crack to enter. This is a new dimension.
There is a reason that the US bans foreign-built mobile robots now.
Some factories already run “lights out” - but there will always be manual tasks… deciding what it is the factories produce in the first place being one example.
Machine vision is literally machine learning/ai.
Where do you think the LLM algorithms came from? It’s just translating language into a easily processable form and that doing a little image processing on that, in a sense.
Token processing and image processing share similar roots, but I’d say each is a specialized branch of a common ancestor rather than either being derived from the other.