For most of November, all I saw were articles about an AI bubble. Now, we have a Santa Claus rally coming. Is there an invisible hand?
Trending Content - Off Topic Forum
| +51 | SpaceX IPO - what are your thoughts | 11 | 35m |
| +28 | Should I use Claude or chatgpt? | 15 | 21h |
| +21 | Funny Things to Say to Interns | 7 | 23h |
| +18 | London's future as a financial centre | 18 | 2d |
| +17 | How did you meet your significant other/partner? | 13 | 1d |
| The Great Debate | 9 | 1d | |
| +7 | E ink/electronic notepads | 4 | 1d |
| +6 | Stay Sharp All Day With Best Modvigil 200 MG Tablet | 0 | 1d |
| +6 | Test post will be deleted automatically. | 11 | 1d |
| +5 | get a model before you model | 3 | 2d |
Career Resources
And Worldcom was "proven" until it turned out it as a fraud.
Neither NVIDIA nor Google are "proven" at anything related to AI. AI is a giant hype machine bubble, waiting to pop, and NVIDIA is going to be the hardest hit.
If Google had gone out and very publicly lit several hundred billion dollars on fire, it would have had massive negative consequences. And that is pretty much what they did, minus the accelerant, and you think it'll be different once that becomes obvious?
"at scale" was the operative word here. I'm aware of these use cases in perhaps smaller, less complex settings but when we're talking about enterprise scaled implementation (which is probably the necessary cog to justify valuations), we're still a ways off in my view.
These implementations aren't costless either. Claude code projects can lead to (I forget the proper technical term, not a coder) code bloat. AI customer service agents have been seen to make costly mistakes for companies, hallucinating discounts (go read about the American airline, can't remember which). Marketing could definitely be interesting, I agree. Not to mention cost of implementation, maintenance and any unforeseen externalities.
I'm by no means anti-AI or massive doubter. I believe this is a transformational technology. But as I said, good technology and irrational exuberance are not mutually exclusive.
And how many mistakes are in there? How much time is spent reviewing the output? How much money has this made you, or whoever you observed?
Can you point to some data here? Because if the revolutionary new use of "AI" is simply "more chatbots" then I hate to tell you this, but there ain't gonna be more disruption. This ship sailed.
And is full of errors. Absolutely fucking chock full of hallucinations. "Research" means understanding something. Anyone can google something, take the first hit that comes up, and claim they've done research. Facebook is full of grandparents who do their research.
Right, but these things are absolute money sinks for the firms that run them, so this won't last. $100 of AI credits buys you how many thousands of dollars of compute?
And I love the laughable "if done right.' So after I burn through 5x as many tokens as I need to figure out the kinks in my AI generated video, I've maybe got a worthwhile ad. To put into a market flooded with ads that will look and sound and feel eerily similar, because they'll all be generated by AI which trained on the same data and thus outputs the same basic shit. Wonderful!
Anyway, you lost the whole thread here when you talked about the subsidized cost and not the actual cost. If I spent $100,000 making an advertisement, I get professionals and their equipment and actual human creativity and at the end of the day I know how much I spent, and how much the next one will cost.
What happens when I'm on a deadline to produce an ad, and suddenly the firm that provides access to their chips sees a spike in activity and shuts me down to favor a better customer? What happens when my budget for the year assumes I'm paying $1/AI credit and suddenly Amazon decides they're not subsidizing my compute usage any more, and jack their prices up to $10/token to reflect their actual cost of compute, plus some profit to start paying back the hundreds of billions of dollars they've spent or lost? Well, now I'm out of business. But at least I got to ask a computer to make me a generic looking video!
Right. Well, this report from June mentions that AI "agents" are successful at single step tasks only half the time. Single step requests!
The revolution is right around the corner, we swear!
What about betting on bots playing poker against other bots? That idea could never be manipulated.
I do think the AI wave is more than LLMs which we use at the moment. A retired VC told a small group of us on the phone about 6 months ago that when the Internet boom first happened, a lot of people thought that was the end point of the technology - it’s going to be the Internet and that’s it. Turns out, there were so many applications and new industries created leading up to today. I think this will happen again.
What were the applications and new industries the internet created? Which of them were unforeseeable in 1995 or whenever? What do you expect LLMs to create?
I'm sorry, but saying "some people were wrong about the internet and its uses" is not at all a justification for "that means AI skeptics are wrong about LLMs."
People managed to monetize the internet in lots of interesting ways, but at the end of the day, what do we use the internet for that is genuinely "new"?
I’m just a real estate finance guy, so I’ll give this a shot. For the Internet, people created companies that could do all kinds of things from e-commerce (like selling dog food) to infrastructure for the internet. Some things were too ahead of their time. But out of it came Search, Social Media, Mobile, Enterprise, Consumer SaaS, Cloud, all the things that built on the next over the 20 years post 1999. Everything that has shaped careers, politics, education, competition, etc.
If we see LLM/Chatgptas the hot thing now (general AI), maybe we see personal/individual AI, Agentic for enterprise, to synthetics/robots, social / politics evolution, techno-religion, AGI, eternal life and all its technological implications (powered by advances in AI, energy, genome, etc).
To say AI, and specifically LLM’s “ain’t worth shit therefore this is hype” is missing the point that this world is changing rapidly.
Whether or not it is only the hyper scaler infrastructure companies making money or other folks, that’s a question. But the cost of doing business is going down. Cost of living is going up for humans (for AI beings, robots it is going down).
Cost of doing business is going down, yet the cost of living as a human is going up relative to income (if we include our observations in our Future of Real Estate discussion: in a low growth world) How can that be?
Does a bear shit in the woods?
The biggest issue I don't think anyone has really raised in this thread are the hardware costs. The internet was able to explode because companies installed miles of fiber and networking gear that was sold off cheaply after the dot com bust, but AI fundamentally doesn't have an analogue. The software is important, sure, but the entire backbone are the GPUs that perform the token calculations, GPUs that burn out or become obsolete pretty quickly. This requires a huge continuous CapEx expense that needs to generate high revenues. And unlike the internet, LLMs aren't marginally free. Once you set up a website, additional views have a marginal cost of essentially zero (in the short term), and while the marginal cost for an LLM to write a response isn't fixed, it's not near zero either.
I think the bubble bursts when there is hard evidence their depreciation schedules don't actually line up with the useful lives of these GPUs. In 1-2 years when these tech companies admit to having to continue to shell out billions for NVIDIA chips there will be a bigger reckoning, I'd imagine. And even if they've found a way to profitably run AI models, the valuations won't be massive multiples, why should they be? A company can only justify huge multiples if there is the promise of massive growth without an equivalent massive influx of CapEx. Tech companies can command large multiples because they're traditionally cap light, but AI can't offer the same investment experience.
There are 5 generation old GPUs in data centers that are still profitable. This entire line of thinking is perpatrated by people who know a little but have zero understanding of the economics of the industry. Just because more powerful chips come out doesn't mean that the old ones instantly become obsolete, you just reallocate them to lower priority or less demanding task. This is what has been done in the industry since it basically became a thing.
Non ea impedit provident quidem atque quo suscipit ut. Qui vel doloremque neque tenetur adipisci quia. Veniam in perspiciatis aspernatur pariatur consequatur temporibus.
Aspernatur eos rerum expedita porro rerum. Dolor odit expedita voluptatibus dicta. Sed quis voluptas et dicta doloribus dicta. Blanditiis cumque incidunt quia saepe eos.
Adipisci excepturi soluta perspiciatis unde dolorem aut est. Quo quam vel sequi distinctio est placeat. Aliquid cum magni quod a vel. Perferendis eum nisi in minima assumenda vel.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...