Am I Being Gaslit by AI?

I feel like I'm being gaslit by AI.

Not by any particular model, but by the conversations surrounding AI in general.

AI has become one of the top topics of discussion in all areas of life. At work, we're all talking about how we can use it to be more productive. With friends, we're talking about the things it lets us do. Online, you're inundated with content about how it's going to change the world and take your job.

It feels like the Kool-Aid has been poured into our water pipes, and we're now as filled with AI propaganda as we are with microplastics.

ChatGPT says this. Claude says that. Just ask ChatGPT to do it. It messed up? Well, you did it wrong. Why didn't you write a 50-page dissertation explaining what it should do? It's on you, really. Oh, you did write all of that? Well, just wait 6 months, and the next model will be even better. And if not, don't worry; the singularity is imminent, and AGI will be here to govern us.

It's suffocating.

Don't get me wrong, I use AI tools daily. At work, they're faster than I am at writing code and searching a large codebase. On personal projects, they help me quickly prototype new ideas. And there are open-ended questions I find LLMs great at helping me work through. It's become a valuable tool in my life.

But that doesn't mean I believe the hype. I see a useful tool that has real impact, but conversations about AI rarely end there. Everyone has accepted the talking points of the AI CEOs, and every conversation about AI is premised on these points being gospel.

You can't say you don't think AI will inevitably get better and take all our jobs because that's a given. Sam Altman says it, and have you seen the progress in the last two years? The line went up, so of course it must keep going up. You can't criticise someone for using AI tools that are not good enough yet because they're getting ahead of you, and you'll be left behind while you complain. You can't point out fundamental flaws in the technology because the AI company says the new model is in the foothills of the singularity or something. Besides, they're valued at a trillion dollars, so what do you know?

It all adds up to feeling like you're being gaslit into not trusting what your brain can see. Those errors and flaws are your lack of foresight and expertise. They don't exist, and even if they did, you're a fool for believing they won't be fixed soon.

But I'm tired of sitting through these conversations. I feel like Ayo Edibiri's t-shirt in that photo above. I want to run away when I get trapped in one of these discussions. If we can't talk about this tech and industry for what it is now, then what's the point?

But unlike the t-shirt, I'd rather not kill myself. So let's have an honest conversation about AI, the hype, the impact, and the lies. If you want to tell me I'm wrong, please do, but tell me I'm wrong based on what both our eyes can see, not some promise by our AI messiahs.

The lie of AGI

Let's get the big one out of the way: generative AI will not lead to AGI.

If we go by the traditional definition of AGI, a type of artificial intelligence that matches or surpasses human capabilities across virtually all cognitive tasks, this technology is fundamentally incapable of it. It can feign intelligence. It can convince many it's alive. But by its nature, there's no reasoning happening here. It's just vastly better tech at giving the most probable output to a given input.

Because of that, an LLM will always be capable of hallucinating. By sheer computing power and time for trial and error, it can accomplish things that would've taken humans longer. But is that intelligence? Is a calculator intelligent because it can complete math faster than you can?

Despite being trained on all of human knowledge, there's just no evidence today that LLMs are as capable as humans in all cognitive tasks. Only on a narrow subset of tasks where their nature is most suitable and output is predictable and testable.

Now lots of people will counter that their conversations with AI are evidence that it's intelligent. But lots of people believe their pets understand every word they say. Doesn't make it true.

Not to mention, the CEOs keep changing the definition of AGI and declaring its arrival suspiciously close to their planned IPOs. When has the Silicon Valley crowd ever led us astray?

I love hanging out with my friends in the metaverse and repaying them for our virtual drinks with bitcoin.

AI can communicate, but is it good?

Communication is the whole thing that made our species and civilisation capable of reaching this point. There's value in being able to condense your thoughts into a distributable form. It can be tiring and difficult, but everything in life worth being good at is.

Many people are now defaulting to AI for all of their communication. Losing any unique voice and diminishing their own thoughts in favour of what a hallucinating algorithm spits out.

That's bad, but it would be less bad if LLMs were actually good at communicating. Most models will spit out unwieldy responses that no human who is good at communicating would ever. They give too much context since they can't know the recipient's full context and the relationship between the people communicating. This leads to paragraphs of text that are a pain for the recipient to read. It can be technically correct but inefficient.

Not to mention the sheer inconsideration of sending someone paragraphs of unverified and unedited AI text and expecting them to read it and make sense of it for you.

These tools are helpful for writing, but it's on you to confirm what it says, edit it, and make it efficient for the person you're communicating with.

I've also seen many people now employing AI agents to run their business. Let's be clear: an AI agent cannot run your business.

Agents have been a useful form that this tech has been put into. The idea of an autonomous machine completing tasks for you is brilliant. But due to the general inefficiency of the way LLMs communicate, it leads to extended conversations that would've been faster if two humans just spoke. I've been stuck in email threads for too long with AI agents that just keep repeating the same 5 paragraphs in different variations.

They're bad communicators, and you're entrusting your business to a tool that is prone to mistakes and can not grasp context you didn't provide. These tools are not ready and only make your business look terrible.

AI answers can never be trusted

Speaking about hallucination, we should all be able to agree that AI answers can never be trusted. If you're asking a question that has an objective factual answer, an LLM can never be a reliable source.

It's faster than finding the real answer yourself, but they're also regularly wrong even if they provide sources. If you're quoting an AI-generated answer to someone, you're just offloading the work of verifying it.

This amplifies some of the worst effects the internet has had on society. We've already struggled with misinformation that's been created and spread by humans. But this is a whole other scale, and it's trusted even more easily when placed at the top of Google results.

At least we used to get warned not to believe everything you see online. But now we're told the AI knows all; of course you should trust it.

This problem cannot be solved by bigger models. AI models are trained on both Wikipedia and Reddit. And they have no inkling of what is truth or opinion or joke or plain lie.

Hurting creatives

Everyone knows this critique. AI makes it harder for real creatives and diminishes the value of their work. The rebuttal is that it's a new tool, just like Photoshop was once a tool. I see some merit in that, but it feels disingenuous not to acknowledge that the harm to creatives is real and will have consequences for them and society in general.

First of all, just because you have a tool doesn't mean you'll be good at using it. It doesn't mean you have good ideas to create with it. You can buy the highest quality ingredients for a terrible chef, and you'll still get literal slop.

The same applies here. Taste and an eye for what's good and what isn't are the real value in any creative field. Not the tools. The tools are there to serve the people who know what's good, not to replace them.

Yes, AI helps you make that little idea you had a reality, but that doesn't mean you're now a film director or a tech genius. You skipped the steps that help you learn what's good and how to be good. Maybe you can still acquire those through AI, but you're not there, and most people never will be. Most people just want to consume good things and maybe have fun creative hobbies on the side.

The idea that everyone will now just create the video game they always imagined or the movie they dreamt of is silly. Most people may have imagined these things but lack a fundamental understanding of what makes a thing good, and their creations are usually terrible due to this, even before AI existed.

Once someone hits that moment where they realise this isn't as easy as just having a brilliant idea, most people give up.

We need creatives, and acting like the harm to them is just natural technological evolution that makes them irrelevant is uninformed. The AI companies are willingly hurting creative industries and trying to convince you that you don't need them. They want you to believe you'll create your own entertainment. But like with many industries in the past (taxis, hotels), these tech companies will lure you in with a promise, capture the industry, make it only profitable for themselves and then charge you more for the product when you have no other choice.

Protecting creatives is important if you care about good art. If we don't, we'll be stuck with slop worse than even the marketing execs have been able to force on us.

Is this just a b2b service?

The AI founders and CEOs promised us a revolution. They promised their AI models would replace every job, solve every problem, and be a part of every aspect of life.

And what we seem to be heading towards now is software services for businesses?

Tools that make your business more efficient but have little impact on people's day-to-day lives. No one's planning a vacation every week that needs an AI agent to use their phone to book it. It's a rare event. And if an AI agent can do it, great, but the average person is not going to pay for that. They might use Booking.com's AI agent to do it. But that still leaves the AI model as a B2B service, not a consumer product.

The valuations of these companies can only be justified if they deliver the world-changing software they promised. Something that actually can do most jobs. Something that can make every industry more efficient. Something that regular people will willingly pay to use daily.

But they're unprofitable. Not even close to profitable. And as they hike prices, consumers aren't paying.

They lied. Or were insanely delusional. And even as AI tools continue to improve, there is no evidence that they will live up to the initial hype. They'll be useful productivity tools. Occasional interfaces for regular people provided by specialised businesses. But that's not what they told us.

Disagree?

Please do. I'd love to learn where I'm wrong. But if you're going to talk to me about AI, have some insightful rebuttal to these points instead of just parroting the talking points you've absorbed. Let's have a nuanced conversation and not just default to our overlords being all-knowing and unquestionable.

If not, then just leave me alone before I actually start wanting to kill myself.