文字稿 🔗
WEBVTT
00:00:00.320 --> 00:00:06.480
Generative AI is extremely dangerous, but not because of what the AI labs want you to believe.
00:00:06.719 --> 00:00:09.519
And they are trying to mislead us on purpose.
00:00:09.759 --> 00:00:18.640
Here I'm trying to describe what problems I currently see with generative AI and LLMs, and why I use them anyway for certain tasks.
00:00:18.879 --> 00:00:23.359
I hear you cry hypocrite right now, and maybe you're right.
00:00:24.160 --> 00:00:32.000
The opinions in this piece might not age very well, so for the record, it's October 2026 as I am writing this.
00:00:32.320 --> 00:00:39.520
Since summer, we have heard stuff like AI agents conspired to hack a company without any human involved.
00:00:39.840 --> 00:00:42.960
AI will become too powerful to control soon.
00:00:43.280 --> 00:00:52.000
We have already built Rokus Basilisk, and there's a chance that AI will wipe out humanity within the next decade.
00:00:52.320 --> 00:00:55.119
But I worry more about something else.
00:00:55.439 --> 00:00:59.759
Not because the above are impossible, I don't think they are.
00:01:00.240 --> 00:01:05.120
And not because I think something even more dangerous is on the horizon.
00:01:05.359 --> 00:01:09.120
Will wipe out humanity is severe enough, thank you.
00:01:09.359 --> 00:01:16.799
I worry about things that are not that bad, but pretty bad anyway, and they are already happening right now.
00:01:17.040 --> 00:01:21.519
Those tools are used for bad stuff a lot, maybe even mostly.
00:01:21.680 --> 00:01:23.519
They are a scammer stream.
00:01:24.000 --> 00:01:28.799
We already learned that we cannot trust emails or text messages before.
00:01:29.120 --> 00:01:43.599
But now we also cannot trust phone calls, video calls, podcasts and videos anymore, and supposedly even letters, and we cannot connect anything to the internet because criminals will hack everything even faster with AI.
00:01:44.079 --> 00:01:47.599
Not that I would have connected my fridge to the internet anyway.
00:01:47.920 --> 00:01:52.560
AI is pitched to companies and executives as a way to replace workers.
00:01:52.799 --> 00:01:56.400
Maybe not now, but definitely in the near future.
00:01:56.799 --> 00:02:06.640
They do that by using content they stole from us, and it does not even have to be true to become an excuse for your company to fire you.
00:02:06.879 --> 00:02:08.879
It only has to be convincing enough.
00:02:09.120 --> 00:02:15.039
Often it's not even important whether the company is more or less efficient after such a measure.
00:02:15.360 --> 00:02:20.080
Some manager met their KPI of increasing AI use and got a bonus.
00:02:20.319 --> 00:02:30.479
And the AI lab has more business customers now and becomes more interesting for investors, even though they lose more money now because every new customer loses money.
00:02:30.719 --> 00:02:39.520
That leads us to the boom is giving more money and more power to people who already have too much
Load full transcript
money and power.
00:02:39.759 --> 00:02:44.879
They have proven that they will not act with the benefit of humanity as a whole in mind.
00:02:45.439 --> 00:02:54.879
Fascism and right-wing nationalism are on the rise, the world is getting more and more divided, and those people are funding this development.
00:02:55.360 --> 00:02:58.319
Yet we are inflating the bubble even further.
00:02:58.479 --> 00:03:09.120
And should it pop hard, it will take down multiple stock markets, retirement funds of millions of people, government savings, and maybe even the US dollar.
00:03:09.759 --> 00:03:16.719
On top of that, we are in the middle of a climate crisis right now, and the current AI boom is making it worse.
00:03:17.039 --> 00:03:29.039
Directly because data centers and chip factories use up water and electricity that we would need elsewhere, and we're switching on old, dirty power plants to satisfy that demand.
00:03:29.439 --> 00:03:41.199
And indirectly, because it takes our attention and money away from everything we could do to make the climate crisis less severe, reducing emissions and becoming more resilient.
00:03:42.080 --> 00:03:52.240
And the same people that are inflating this bubble right now and that will probably cause it to pop will get richer when it does, while the rest of us will end up poorer.
00:03:52.560 --> 00:03:59.520
Maybe some companies will go bankrupt, and maybe some rich people will lose almost everything too.
00:03:59.759 --> 00:04:10.879
But for most of them, even if they lose a lot, they'll still have enough to buy when there's blood running in the streets, while we will be the ones bleeding.
00:04:11.280 --> 00:04:17.920
With every crisis, the divide between the ultra-rich and everyone else is only getting bigger.
00:04:18.160 --> 00:04:30.959
If you already follow me on social media, you have read posts where I am extremely skeptical about AI, like this text here, and others where I write about how I use it.
00:04:31.199 --> 00:04:32.480
What's up with that?
00:04:32.720 --> 00:04:37.839
I think it is very hard to completely avoid using generative AI at the moment.
00:04:38.240 --> 00:04:45.759
Yes, some people are actively avoiding it, and for some it even became part of their personal brand.
00:04:46.079 --> 00:04:51.120
But I'm not convinced that this can work for everyone, at least not right now.
00:04:51.680 --> 00:04:54.319
It's a bit like the tragedy of the commons.
00:04:54.720 --> 00:05:02.240
Using it is bad for everyone, but not using it puts us personally at a disadvantage.
00:05:02.560 --> 00:05:11.120
And it is unfortunately also very useful for some tasks, and there are some things that I could not do without those agents right now.
00:05:11.839 --> 00:05:21.040
Also, some folks will happily use AI, others are forced by their employers or pressured by their team to use it.
00:05:21.279 --> 00:05:32.000
And since I, as a technical coach and trainer, work on helping teams and individuals to get better at what they do, I need to know the tools they are using.
00:05:32.240 --> 00:05:37.680
I think the world would be better off when we did not have the bubble that exists right now.
00:05:37.920 --> 00:05:46.160
Some problems I have with generative AI might be fixable by smaller, more local models, but not all of them.
00:05:46.399 --> 00:05:51.279
And definitely not in this hypercapitalist environment we are in right now.
00:05:51.519 --> 00:06:04.800
I don't really have a solution, but I think we might already be a little bit better off when more people would be more skeptical and chose more consciously what to use AI for and where to avoid it.
00:06:06.959 --> 00:06:09.759
The next essay will be in my usual tone again.
00:06:10.000 --> 00:06:10.959
Hopefully.