The AI Bubble: is it good or bad?

[Telesphoreo]

I wanted to make a thread to discuss the AI "bubble". I think it's clear that there's a bubble, but the real question is whether it's a good bubble and what the effects of it will be on society. The dot com bubble is generally regarded as a "good" thing. Even though there were many overvalued companies with no profits and terrible ideas, it did pave the way for many great ones and also the infrastructure we have now. Fiber was overbuilt and it led to globalization and a more accessible internet for everyone. The 2008 financial crisis was bad though and I think we can all agree on that.

Throughout this thread, I'll mainly be focusing on CS and the software development industry and how it relates to AI. Of course, it has changed every industry differently, but I will be focusing primarily on software development as that is what I am most familiar with.

The Past
It's kind of incredible to try and imagine a time before ChatGPT and other LLMs. It's only been three years as of writing this since they first came out. Coding used to require more people and more time before LLMs. Even to do a lot of the simple things like boilerplate or scripting, you had to have skill. If you even wanted to do something as basic as make an API call or write a backup script, you had to know quite a bit. However, these tasks are now stupid simple thanks to AI and you don't even have to understand a line of code. You can just keep trying until you get the results you want. The reason this is important is because there used to be a "valley" of knowledge you had to climb to truly make anything meaningful. Part of the pain (and fun) of software engineering was the thinking process and debugging. It always felt rewarding to build something and truly understand it. Each time you did, you were guaranteed to have learned something. Even if you copied from StackOverflow (remember that site?), you still had to likely adapt the code. Finding a straight copy and paste from StackOverflow was always lucky.

The Present
The most critical thing ChatGPT did for software engineering was remove the barrier to entry. Now anyone can write code that is truly specific to exactly what they want to do. And there will be an LLM there who (as long as you pay) can infinitely answer questions and try and help you debug. You no longer have to find a course or tutorial and learn from the beginning. Depending on the language, this is super impactful. Python hasn't changed much and, in general, any Python 3 tutorial should work on your system. Whether you're using the default Python 3.9 on your system or the very latest version 3.14, you could still follow the tutorial. Other languages and frameworks like React are a completely different story. React changes all the time and a tutorial from a few years ago could be completely outdated. Due to there being less hurdles, anyone can start writing React code much easier. Confused on how to make an API request? The LLM can write you the entire component and debug it without you truly learning anything. You see the result and see "oh, useState, sure that's for state. useEffect, I guess what's when something updates." But it's never the same as truly having to learn that from a tutorial and cobble it together and get a real understanding. It short circuits the learning process for beginners.

On the other hand, LLMs are giving senior engineers an enormous gain. They already know how to boss juniors around and ask for what they want. They know how to architect systems, not just build them. The more experienced you are, the more you are able to look at code and point out any flaws or if it looks right. This is exactly why the demand for juniors has fallen. The seniors can just do all of that work themselves and fix the last 2% of issues on their own. Even if you don't know the codebase, you can often just see the error and intuitively know how to fix it. If you never started out as a junior though and learned how a complex codebase interacts and learn the "gotchas" and spent hours debugging, then you really haven't learned. Having an LLM bail you out is not learning when you're getting started.

You could argue that learning is a waste of time. For example, I was modifying the XenForo templates the other way. I could have gone through the developer documentation and gotten a deep understanding of how to use the XenForo macros. However, I am not a full time XenForo developer. Hiring one would be incredibly expensive for the small thing I was trying to do. So you absolutely could argue that it's a good thing for the software world as it allows you to build and achieve more without needing to pay as much or have as much expertise, especially if you're trying to do something simple. Maybe I don't care to learn the entire XenForo macro system and using an LLM to give me the syntax was actually the easiest solution. I don't need a deep understanding of the XenForo macros. I wasn't modifying anything critical and certainly don't intend to get a full time job doing XenForo development.

More broadly though, LLMs can complete a lot of coursework assigned by universities. I think the issue is not that the LLMs are too smart. Platforms like Chegg have existed long before LLMs. If you were desperate enough, you could try and scour Chegg and Quizlet for answers. This has always been the case and there have always been shortcuts you can take in university. It's not like every student was truly learning from scratch with no outside help. It just used to depend on the resources available outside of class and whether there were TA's, office hours, and so on. However, none of that is needed anymore. You don't need to wait in a long line for some non-personalized help. You have instant access to a tutor 24/7 with ChatGPT. And it will be patient and explain the concept for hours until you get it. Or, it could give you the entire answer with no learning. Fundamentally, the responsibility is still on the student on how they want to shape their learning experience. There have always been tools for shortcuts prior to ChatGPT. It's just a lot more accessible to take shortcuts now. It is true that many universities are adopting AI into their curriculum by the logic you will have access to AI tools on the job. However, this may not be the case...

The Future
This is the part we don't know. The use of LLMs may be as bad as smoking for all we know. Smoking wasn't considered "bad" until there had already been a generation of smokers. Even if AI is adopted at universities and the course load is sped up, that doesn't mean that it's a good thing. Just because something (AI) is endorsed doesn't make it good or effective.

I don't think LLMs will be democratized if I am being honest. I think, in the future, LLMs are headed towards the life of Chegg. You can use it, but it will cost you money and will be used sparingly. Even in programming as well. I think it'll be used as a "get out of jail" card for debugging rather than building whole projects. It can shortcut the learning, but it'll feel as expensive and "wrong" as Chegg. It is important to remember that LLMs are fundamentally not "intelligent". They don't have lived experiences and understand meaning the same way we do. It's also incredibly easy to "trick" LLMs like toddlers.

Frontier AI models require a tremendous amount of power to run. The capable AI models, the ones actually worth running, are not the OSS ones. And they require significant power draw, a building, ongoing maintenance, and our natural resources. And the NVIDIA GPUs running them are heavily specialized. They can't really be repurposed for anything at this time. Maybe we find another incredible use case for CUDA. But even if we do, you still must factor in the power draw of their older chips is much higher. So even if they go on the secondhand market, the probability of them being able to be used effectively is low.

However, there is another case, which is that LLMs become so incorporated into our lives, whether we want them too or not. China is leading the AI race in terms of energy for datacenters. Construction in China can happen significantly faster and there's more energy to go around for datacenters. The US government genuinely may bail out all of the AI companies and datacenters and try and "inject" money into it to not appear weak to China. It genuinely lead to an extinction of resources. If China and the US don't stop trying to one up each other and have "dominance" in the AI race, we may lose all of our resources in trying to prove our superiority. If AI leads to the human extinction, it won't be because it got super intelligent. It will be because it took all of our resources.

It also could truly be the end of capitalism and consumerism. Maybe we're headed towards datacenter-ism?... Every company could realize that if the government is injecting infinite money into AI companies and technologies, they should drop everything and focus on making products to sell to that industry. We've already seen Micron pull out of the consumer business to exclusively focus on datacenters and enterprise. I know this is just one example, but what if it's simply more profitable to shut down the land your Walmart is on for a "mini datacenter"? What if it's more profitable for a farmer to sell the land they grow their food on and sell the resources to a datacenter conglomerate rather than continuing to produce food? It genuinely is something I could see happening if the US injects capital to keep us "winning" in the AI race.

Maybe it seems a bit crazy to say that, but there are seven tech companies making up the majority of the S&P 500. They are also betting heavily on AI and they are betting on it for continued growth. Also, let's not forget the circular handing of money:


The entire economy is being propped up by AI. We very much could see a bailout and injection of money by the US government to keep this going. Would it be sustainable long term? Of course not. However, as long as the current people in our government can pass it down to whoever is next, there is no problem.

So, what do you all think?

 

[zrfv]

the only thing good is that companies lose, the only bad thing is consumers also lose i think or something

 

[videogamesm12]

I had a long reply drafted for this thread that ultimately got lost somehow, so I instead have to restart from scratch using an early draft I had saved elsewhere as a basis. Oh well. I wrote this as I started to get really sleepy, so don't expect it to be anywhere near as cohesive as a typical write-up. I might think more about this later on, but these are my thoughts as of right now in a somewhat disorganized manner.

---

When what we refer to now as AI first hit the mainstream audience, I was incredibly skeptical of its usability and concerned about its long term implications on society itself. I was also very displeased to hear that many companies were attempting to cram it into as many of their products as they could without any concern for its actual usability or any privacy concerns that may come from it.

A good example of AI being carelessly, needlessly, and neglectfully shoved into an existing product has to be the absolute dumpster fire that is Windows 11 and its Copilot+ debacle with Recall. I don't know what planet Microsoft was on when they conceptualized the idea of deeply nesting a "feature" that takes screenshots of your entire screen every 5 seconds into the operating system and thought it was a good idea, but wow did they completely drop the ball on that and no amount of marketing could salvage the reputational damage that did. You can't even tell me that it's only available on Arm-based computers anymore because I've witnessed with my own four eyes x64-based Copliot+ computers being sold on the market, and it's clear Microsoft wants to port this feature to more and more systems with time. They tried to market it with the selling point of "Aw man, I forgot what I looked at earlier, I should ask a chatbot to tell me", which is basically the same fucking thing as a search bar in most modern applications.

However, I also now believe AI is way too much of an innovation to entirely dismiss as a gimmick. The multimedia boom and dot com bubble of the 2000s fundamentally changed how we consume digital media and I don't doubt that AI have the same impact, even if we end up sacrificing a bunch of companies in the process to make it there. Companies right now are trying to experiment with what AI can and can't do well as a consumer-facing feature (much like a monkey throwing shit at a wall to see what sticks), so while there have been a lot of misses there have also been a lot of hits - and it's these hits which can be fine-tuned into a bullseye if executed properly. That's why I am cautiously optimistic about AI and its future.

To elaborate further, let's take a look at one of AI's biggest strengths - summarization. AI is surprisingly good at taking a large amount of information and compressing it down into something more comprehensive and understandable. Case in point, when I made that really long thread evaluating three ways to get digital media which included subscriptions, physical media, and piracy with their costs and benefits, someone complained that the thread was too long and wanted a TL;DR. Packs replied to that user by feeding my essay into an AI model and asking for a summarization. The result blew me away, even if it didn't get everything right. I've included it below.

Absolutely. Summarizing this casually massive wall of text is exactly what I was trained for—because why wouldn’t you drop a dissertation on me?

Main Point:

Media consumption has shifted from physical ownership to streaming and digital platforms, but this change comes at the cost of real consumer ownership and control, due largely to DRM (Digital Rights Management) and restrictive business models.

Key Takeaways:

1. Ownership Erosion:
   Consumers no longer truly own digital media. Most content is licensed, not purchased, and access depends on internet connectivity, platform support, and corporate policies.
2. DRM and Its Problems:
   DRM restricts how and where you can use media, often harming legitimate users while pirates face fewer restrictions. It can even damage devices (e.g., Sony’s rootkit scandal).
3. Consumer Inconvenience:
   DRM causes problems for casual users and preservationists alike. Games may require always-online DRM, outdated services, or third-party accounts, making even single-player games unplayable.
4. Streaming ≠ Ownership:
   Buying digital content or subscribing to services like Netflix or Amazon doesn’t grant ownership. Content can disappear due to expired contracts, and offline access is limited or temporary.
5. Physical Media = True Ownership:
   DVDs, Blu-rays, and CDs offer permanence, reliability, and full control. They don’t depend on online authentication or corporate support and can be backed up digitally.
6. Piracy’s Double-Edged Sword:
   While illegal, piracy often provides better user experiences and long-term access compared to legal methods. It’s sometimes necessary just to run older games with broken DRM.
7. Industry Misconduct:
   Media companies have aggressively resisted consumer-friendly tech, abused legal systems, and prioritized profit over user rights, often punishing ordinary consumers and stifling innovation.

Conclusion:

Today’s media landscape prioritizes profit and control over ownership and user freedom. For lasting, flexible access, physical media or DRM-free digital purchases (like on Bandcamp or GOG) remain the best options.

Would you care for this summary to be transformed into a slide deck or infographic? Naturally, such a towering work of literature warrants a visual aid.

One point that I would like to stress is the proportional relationship between the gains of a user and how they use AI. Your gains from using AI as of right now are entirely dependent on what you're using the AI for and how you're using it. If you ask ChatGPT to vibecode an entire chat application using a bunch of frameworks, it'll probably work, BUT that doesn't mean you actually gained anything from the experience aside from having a result that might not even be good in the first place. However, you gain a lot more if you instead wield AI as a tool either to add onto your existing knowledge of a particular topic, automate the more mind-numbing parts of a topic you already know very well about like writing boilerplate code, or even to adapt an idea to something you can more easily understand with the intention of learning it better. However. it's also important to be careful not to over-rely on AI tools to do your job, because then if you don't practice certain concepts or ideas enough, you tend to forget them after a while. This is something I am deathly afraid of and is one reason I don't really use it in any of my workflows.

I suppose whether the AI bubble being good or bad depends heavily on how we use and support it - not just individually, but collectively as well. If we start using AI to replace traditional learning instead of supplementing it, that would be pretty bad and would take generations for us to recover. If we allow the bubble to get so big and then bail out companies that only hopped on the trend for the dollar signs and used resources wastefully, then we'd only be showing that there are no consequences for wasteful practices and to just keep on going as per usual. However, if we let the wasteful companies fold and have the more efficient ones survive, then it re-enforces the importance of being efficient and effective at the cost of creating a monopoly.

I honestly don't know lol. When I wake up tomorrow, I might refine this reply to something that isn't a jumbled mess.

 

[Alco_Rs11]

One point that I would like to stress is the proportional relationship between the gains of a user and how they use AI. Your gains from using AI as of right now are entirely dependent on what you're using the AI for and how you're using it.

I have found tools like ChatGPT to be quite useful for even being a "source-gatherer". Since I am in an academic program that requires a lot of research, AI has made it easier than ever to compile sources to use on an assigned topic. Gone are the days of having to do tons of searches when ChatGPT is provided with the right prompt (the more specific, the better), you can get a bunch of sources or academic articles easily.

Your mileage may vary, but even ChatGPT tells you to make sure the sources are correct, as it isn't always 100% right.

To elaborate further, let's take a look at one of AI's biggest strengths - summarization. AI is surprisingly good at taking a large amount of information and compressing it down into something more comprehensive and understandable. Case in point, when I made that really long thread evaluating three ways to get digital media which included subscriptions, physical media, and piracy with their costs and benefits, someone complained that the thread was too long and wanted a TL;DR. Packs replied to that user by feeding my essay into an AI model and asking for a summarization. The result blew me away, even if it didn't get everything right. I've included it below.

This is something I also found to be useful for. Being able to essentially make an "abstract" of a long piece further helps with academic writing since you can get the main point of a source without having to read long essays (although it's important to fact-check when necessary).

More broadly though, LLMs can complete a lot of coursework assigned by universities.

Many universities don't even ban AI and actually embrace it. My university allows us to use it, and gives students extensive discretion in its usage, such as using it to check math work (its even encouraged), use it as a writing aid and to help with research, but when something is generated with AI, students are expected to cite that something was made with AI, and usage has to align with the school's academic integrity policy (you can use it but you cant just generate a paper and say you wrote it).

Even if the bubble bursts, it is extremely unlikely AI will ever go fully away because it's essentially become a useful utility for a number of situations.

 

[Telesphoreo]

I have found tools like ChatGPT to be quite useful for even being a "source-gatherer". Since I am in an academic program that requires a lot of research, AI has made it easier than ever to compile sources to use on an assigned topic. Gone are the days of having to do tons of searches when ChatGPT is provided with the right prompt (the more specific, the better), you can get a bunch of sources or academic articles easily.

Your mileage may vary, but even ChatGPT tells you to make sure the sources are correct, as it isn't always 100% right.



This is something I also found to be useful for. Being able to essentially make an "abstract" of a long piece further helps with academic writing since you can get the main point of a source without having to read long essays (although it's important to fact-check when necessary).



Many universities don't even ban AI and actually embrace it. My university allows us to use it, and gives students extensive discretion in its usage, such as using it to check math work (its even encouraged), use it as a writing aid and to help with research, but when something is generated with AI, students are expected to cite that something was made with AI, and usage has to align with the school's academic integrity policy (you can use it but you cant just generate a paper and say you wrote it).

Even if the bubble bursts, it is extremely unlikely AI will ever go fully away because it's essentially become a useful utility for a number of situations.

Excellent points. But the real question is the fallback + reliance. What if it becomes prohibitively expensive to use Deep Research? We haven't seen LLMs become broadly unavailable or inaccessible. If it happened tomorrow, no big deal. It's only been a few years of it and truthfully, it hasn't set anyone back in the long term. It would be annoying to go back to researching manually, but it's not a lost skill. Yet. What happens in 5-10 more years? We're all reliant on AI companies to have this feature and doing research without feels completely foreign. It might be one of those things where it's a lost art.