When Will The AI Bubble Burst?
All this AI crap is looking like it's charting a path very similar to the early-2000s dot com bubble. A few companies that are/were responsible for the fundamental hardware infrastructure of the technology being bid up to absurd valuations (Nvidia vs Cisco). VC people chucking money at anything that proports to incorporate AI into their products (in 2000 it was anything on the Internet). Companies just shoving AI into anything as a marketing gimmick without putting any thought into whether it actually belongs there (in 2000 is was just sticking anything on the Internet). Companies releasing buggy broken garbage with little actual tangable value just to get on the hype train. The one big difference is the fact that the dot com bubble was fueled by mostly startups with little established revenue whereas the AI bubble has been been driven mostly by large established players that have tons of cash from other ventures. I'm speculating that the AI bubble will take longer to burst because of this. When will investors finally realize that AI at this point is mostly just marketing bullshit and hype designed to steal their money? TC 350k
hallucinations are only a issue because people are expecting sync conversations cause of the stage set by chatgpt. if either async was to be the norm or if throughput was 1000 tokens per second, hallucinations won't be a problem. Both of these are achievable.
Huh? How would async solve hallucinations?
There's research showing that if you give the model more time to "contemplate" on their answers, hallucinations drop. It's similar to how we can ask ask the model today to rethink their answer, and the next answer is more likely to be accurate
It is not going to burst entirely but only the big players will survive. The hundreds of current AI startups will all disappear over time. They are all features, not companies.
So basically the same as the dot com crash.
Most of the current AI startups are glorified ChatGPT wrappers. The real fun will be when a malicious actor poisons a HuggingFace foundation model.
The Nvidia bubble will burst long before AI runs out of steam as the big players move to custom HW
It's also going to hit a limit where the ai is going to be trained off of ai generated content. I just can't see it not having any defects. At some point new models can't exceed old models and it will all come crashing down.
So The 5000 Fingers of Doctor T is both a movie and a training model
It depends what you and investors mean by AI. If you mean cheesy chatbots and LLMs I believe in a year at max most people will be disillusioned with them, but if you mean neural networks in general and their evolution it will keep on going for a while. Nice comparison tho, now that you laid it out this way I can agree it is a dotcom v2. We still have internet and internet is still one of the most lucrative tech of our times.
This is the “yeah, but Blockchain” argument that has a nugget of merit and, dull surprise, similar energy demand concerns.
It’s even worse than the dot com bubble because the Internet was a really useful technology and so, the survivors actually flourished. AI is much limited in usefulness in comparison. Big tech has been using AI in a lot of places (to dubious results) and will continue to use them with iterative improvements. All startups hoping to build businesses using AI will go bust. Most smaller tech and non tech companies will realize there isn’t much to be gained from AI, especially in comparison to the investment required.
AGI is humanity’s final invention and the inventor and controller makes themselves the eternal god emperor. This bubble never pops. Resources will pour into this race until AGI is achieved, after which money will become meaningless.
When GPT-5 is released or shortly thereafter, marking the peak euphoria followed by the realization the improvement was incremental and not exponential as every AI bro hoped it would be. When this bubble pops, it will be unlike anything ever seen. Even 2008 GFC will dwarf in comparison. AI is the dominant bubble but it is far from being the only one. Garbage like crypto and so many inflated assets including meme stocks are on their own epic bubbles as well. Internet/Social Media is a big amplifying factor today which was missing in earlier bubbles. It is like international travel for spread of a new pathogen waiting to creating havoc across the entire global system.
The improvement IS exponential, just inversely.
If you think AI is a bubble, then you are in denial.
Hello, 2022 crypto stan 👋
At the moment, the companies that are riding the current AI hype train command absurd valuations while making no money on AI and having 0 idea how to monetize it. The only companies making actual significant revenue on AI are chip companies that sell their chips to companies that make no money using them. AI will likely eventually be a significant revenue driver in the decades to come, just like the Internet eventually became a revenue behemoth. However, if this happens it will be many years from now. The market is currently being far too optimistic when it comes to expecting profits from these companies so quickly. It's going to crash in the short to medium term.
We’re going to hit a limit where GenAI can only scale for a limited set of applications where hallucinations don’t matter. Then the built up GPU compute will be abandoned (like how the dot com bust left dark fiber in its wake) to be hoovered by scam bot farms. The craptastic “Why don’t people like this picture” meme garbage polluting Facebook is only the beginning.
Google and Microsoft already shoved it into search, and it's pretty useless there. They've already run out of ideas. Investors just haven't realized it yet.
Monetized SEO started Google on the path of useless search