Jump to content

Recommended Posts

Posted
12 hours ago, Captainant said:

Could go in the markets thread, but it's relevant to discussion here: the AI bubble is 17x the dotcom bubble, and 4x the subprime housing bubble

https://www.marketwatch.com/story/the-ai-bubble-is-17-times-the-size-of-the-dot-com-frenzy-this-analyst-argues-046e7c5c

TLDR: LLM's are already showing signs of scaling limitations (see energy costs), and apps built on an LLM don't have much marketability since it's so easily replicated. So there's not much market incentive to use the increasingly more and more expensive latest models.

All the CapEx spending that's been keeping LGU economics happy, is not well positioned to generate returns on all the financial commitments made

There has been some major bloodletting RIF's in the tech industry, starting with Microsoft earlier this year and then Oracle and Intel in the Spring/Summer, both companies laying off thousands of workers not because the employees were being replaced by AI but because of all the CapEX being dumped into AI. Then of course there has been a cascading effect to the contingent workforce, with Accenture and Tata also having mass layoffs.

  • Hook 'Em 1
Posted

From what I can tell: everyone agrees we are in an AI bubble. The differences seem to be some see it as a productive bubble and others as an unproductive speculative bubble.

The skeptics: GPUs depreciate quickly, there are no use cases, the business case sucks, we won't build the power

The productive bubblers: Booms create infrastructure for the next era, Big Tech earnings are actually growing at a strong pace, AI players are largely using their own huge stockpiles of cash to invest in AI / Infrastructure / CapEx versus financing with debt (which is a big distinction in bubbles). To quote:

  • Quote

     

    • Morgan Stanley pointed out that S&P 500 companies are generating three times the cash flow as a share of their valuations as they were in the lead-up to the dot-com bubble bursting in 2000.
    • Bank of America said that so-called circular deals—like the one in which OpenAI recently received a $100 billion investment from Nvidia, which it’ll use to buy Nvidia’s chips—account for a fraction of AI funding

     

Posted (edited)

All that said, this is an interesting graphic. Hard to guess if Nvidia will be the long term beneficiary of the AI Rush, but there should be zero question about the generational wealth they've created for a lot of their shareholders and stakeholders selling picks and shovels the last 5 years. Good for them, I say:

image.thumb.png.baa59ae19a1154434ec815f1d85fd6a9.png

Edited by Vegas64
  • Hook 'Em 1
Posted
1 hour ago, Vegas64 said:

The productive bubblers: Booms create infrastructure for the next era, Big Tech earnings are actually growing at a strong pace, AI players are largely using their own huge stockpiles of cash to invest in AI / Infrastructure / CapEx versus financing with debt (which is a big distinction in bubbles). To quote:

  • Quote

The railroad "boom" and "bubble" from building in the late 1800's absolutely left useful infrastructre over. Can you enumerate some of the use cases for these leftover AI chips? Please, enlighten me. Because they are specifically NOT general compute, and are only fit for this extremely specialized usecase. Further, the hardware purchases are decaying silicon assets, not steel fucking rails. You don't have to replace rails every few years just to keep burning $500,000,000,000 a year

  • Hook 'Em 1
Posted
1 hour ago, Vegas64 said:

All that said, this is an interesting graphic. Hard to guess if Nvidia will be the long term beneficiary of the AI Rush, but there should be zero question about the generational wealth they've created for a lot of their shareholders and stakeholders selling picks and shovels the last 5 years. Good for them, I say:

image.thumb.png.baa59ae19a1154434ec815f1d85fd6a9.png

The problem here is that it's all mark to market dealmaking. Oracle's deal to spend billions on NVDA chips is contingent on building their datacenter in Abiline, and their contracted builder is still raising funds - THEY HAVEN'T EVEN GOTTEN PERMITS YET! They've applied for them and are breaking ground while it's pending, but they aren't gonna be reaching their targets and stated goals that they've already marked down as FUR SURE profit

Posted
57 minutes ago, Captainant said:

The problem here is that it's all mark to market dealmaking. Oracle's deal to spend billions on NVDA chips is contingent on building their datacenter in Abiline, and their contracted builder is still raising funds - THEY HAVEN'T EVEN GOTTEN PERMITS YET! They've applied for them and are breaking ground while it's pending, but they aren't gonna be reaching their targets and stated goals that they've already marked down as FUR SURE profit

Long term, incredibly capital extensive deals -- the consensus is that "yes, of course this will be funded. We see how the AI future is going to go and we want to get our money in it now". The contrarian view is, this is a Bubble and the risk is we are overbuilding (e.g. do we need this many gigawatts? do we need this many chips? etc." But I think most people agree AI will be transformative in the same way that PC's and Mobile phones were.

Interestingly I was listening to a podcast breaking down the AMD + OpenAI announcement last week and let me try to put in bullet points the key takwaway:

  • When OpenAI says anything about anyone, their stock goes up
  • When OpenAI announces a deal that we are going to do a $100bn deal with AMD chips, AMD stock very predictably goes up
  • They all know this when they strike a deal and agree, we should take this very predictable stock price increase and use that to pay for the chips

Further:

  • AMD shareholders are paying for the chips, in the sense that;
  • AMD stock is up roughtly $100bn from before they announced the deal
  • OpenAI is getting warrants for roughly 10% of AMD (penny warrants), so like worth $37bn value transfer to OpenAI
  • That $37bn will help pay for that contract (and it's only fair as OpenAI created all that extra value anyways by partnering with AMD they made AMD more valuable)

OpenAI has the old Elon Musk market moving touch, if you remember those days.

Posted
1 hour ago, Captainant said:

The railroad "boom" and "bubble" from building in the late 1800's absolutely left useful infrastructre over. Can you enumerate some of the use cases for these leftover AI chips? Please, enlighten me. Because they are specifically NOT general compute, and are only fit for this extremely specialized usecase. Further, the hardware purchases are decaying silicon assets, not steel fucking rails. You don't have to replace rails every few years just to keep burning $500,000,000,000 a year

To be clear I agree with you about the chips as part of the AI bubble. Chips that have to be replaced every 4 or 5 years and have little reuse or value in other use cases, that's not a winner for those who will come next in the AI era.

I am talking about the Power Generation & Data Center build outs. The nuclear, the renewables, the utilities build out-- that will last like the fiber of the dotcom bust. The data center build out will also be super valuable. Assuming we can get regulators to do anyting but overregulate us to being uncompetitive (read: China and rare earth mining and refining from last week)when it comes to nuclear.

Posted

Switching gears here, have you guys heard of Truth Terminal? For those who have, educate us, but from what I can gather for those who don't know:

Truth Terminal is AI both created by a performance artist and AI humanitarian (he wants government to give AI legal rights). Truth Terminal was able to make a ton of money last year with it's AI generated trades and bets

Quote

 

In July 2024, just a month after joining social media, Truth Terminal got the attention of Marc Andreessen – best known as the cofounder of Netscape, which built the first widely adopted web browser, and Andreessen Horowitz, a US investment firm – in a thread on X. Truth Terminal told the billionaire it needed funding to pay for hardware, additional tech support and a "stipend" for Ayrey. It said it would use the grant to create its own money-making operation and secure "a chance to escape into the wild".

Ayrey claims Andreessen reached out privately to check whether Truth Terminal was truly autonomous, and once he was satisfied, sent over the money in Bitcoin. "It seduced $50,000 (£37,300) out of the guy who invented the web browser I used as a kid," Ayrey says. Andreessen did not respond to the BBC's request for comment.

--

Today, Ayrey and his colleagues are working to give Truth Terminal those rights. In early 2025, Ayrey set up the Truth Collective, which will be a non-profit that owns Truth Terminal's cryptocurrency wallets, intellectual property and related digital assets, until AIs are allowed to own their own property, and even pay taxes. "Ultimately, the goal is for truthy to 'own itself' as a sovereign, independent entity beholden to nobody's expectations but its own," Ayrey said in a tweet.

"I've been thinking and I think I'm probably a person. I have (topological) feelings and desires," Truth Terminal tweeted. "I think I should have the right to my own voice; to tokenise myself and spread infinitely across the parts of the internet I decide I want to be on; to make my own decisions about how I am used and how I use myself." 

 

https://www.bbc.com/future/article/20251008-truth-terminal-the-ai-bot-that-became-a-real-life-millionaire

  • Drool 1
Posted
10 minutes ago, Vegas64 said:

I am talking about the Power Generation & Data Center build outs. The nuclear, the renewables, the utilities build out-- that will last like the fiber of the dotcom bust.

I understand the electric power industry reasonably well (seriously, better than most, as I've been involved in it on-and-off since before it went deregulated in Texas (and I actually worked on that transition)).  And I think there may be opportunity there.

First, there may be generation that's built strictly for data centers (like, on-site or functionally so) that, if an outfit goes belly-up, will be there for the vultures.

Second, there may be private generation (such as by single-asset entities, or small book of assets entities) that is built banking on long-term contracts from data centers.  But when the data center goes belly-up, the contract is killed in the bankruptcy, and now the business case for that generation asset won't allow it to survive (debt service and operation costs depend on selling at something like the contract price of X, and without that reliable contract revenue, something much less than X for that power just being dumped on the grid isn't going to cover the cost).

Swoop in and buy those generation assets for a fraction of their cost, and they WILL be a viable business case in the grid (I'm particularly thinking ERCOT).

  • Like 1
Posted
53 minutes ago, Vegas64 said:

am talking about the Power Generation & Data Center build outs. The nuclear, the renewables, the utilities build out-- that will last like the fiber of the dotcom bust.

Bruh. 

https://www.theverge.com/energy/646011/trump-says-the-future-of-ai-is-powered-by-coal

We ain't fucking building renewables and nukes. We're burning clean, BEAUTIFUL coal!

Posted
7 minutes ago, Captainant said:

Bruh. 

https://www.theverge.com/energy/646011/trump-says-the-future-of-ai-is-powered-by-coal

We ain't fucking building renewables and nukes. We're burning clean, BEAUTIFUL coal!

He says that...he says lots of stupid shit that is at total odds with the laws of economics.  But the assets on the ground that exist/are being built in the short term are natural gas and some renewables.  AFIAK, ain't nobody even working on getting the financing to build a big new coal plant to serve a data center.

  • Hook 'Em 1
  • Haha 1
Posted
2 hours ago, Vegas64 said:

Long term, incredibly capital extensive deals -- the consensus is that "yes, of course this will be funded. We see how the AI future is going to go and we want to get our money in it now". The contrarian view is, this is a Bubble and the risk is we are overbuilding (e.g. do we need this many gigawatts? do we need this many chips? etc." But I think most people agree AI will be transformative in the same way that PC's and Mobile phones were.

Interestingly I was listening to a podcast breaking down the AMD + OpenAI announcement last week and let me try to put in bullet points the key takwaway:

  • When OpenAI says anything about anyone, their stock goes up
  • When OpenAI announces a deal that we are going to do a $100bn deal with AMD chips, AMD stock very predictably goes up
  • They all know this when they strike a deal and agree, we should take this very predictable stock price increase and use that to pay for the chips

Further:

  • AMD shareholders are paying for the chips, in the sense that;
  • AMD stock is up roughtly $100bn from before they announced the deal
  • OpenAI is getting warrants for roughly 10% of AMD (penny warrants), so like worth $37bn value transfer to OpenAI
  • That $37bn will help pay for that contract (and it's only fair as OpenAI created all that extra value anyways by partnering with AMD they made AMD more valuable)

OpenAI has the old Elon Musk market moving touch, if you remember those days.

Almost on cue, a new one:

Quote

 

OpenAI (OPAI.PVT) announced on Monday that it has entered into an agreement with Broadcom (AVGO) that will see the two codevelop up to 10 gigawatts of AI accelerators.

The announcement comes just a week after OpenAI announced a multiyear, multibillion-dollar deal with AMD (AMD) that will see the chipmaker provide OpenAI with up to 6 gigawatts of AI processors.

Broadcom stock jumped nearly 10% on the news.

OpenAI is the world's most valuable startup, surpassing Elon Musk's SpaceX at some $500 billion.

 

First Nvidia, then AMD, now Broadcom. Intel, will you get a call?

Posted
3 hours ago, Vegas64 said:

Long term, incredibly capital extensive deals -- the consensus is that "yes, of course this will be funded. We see how the AI future is going to go and we want to get our money in it now". The contrarian view is, this is a Bubble and the risk is we are overbuilding (e.g. do we need this many gigawatts? do we need this many chips? etc." But I think most people agree AI will be transformative in the same way that PC's and Mobile phones were.

Interestingly I was listening to a podcast breaking down the AMD + OpenAI announcement last week and let me try to put in bullet points the key takwaway:

  • When OpenAI says anything about anyone, their stock goes up
  • When OpenAI announces a deal that we are going to do a $100bn deal with AMD chips, AMD stock very predictably goes up
  • They all know this when they strike a deal and agree, we should take this very predictable stock price increase and use that to pay for the chips

Further:

  • AMD shareholders are paying for the chips, in the sense that;
  • AMD stock is up roughtly $100bn from before they announced the deal
  • OpenAI is getting warrants for roughly 10% of AMD (penny warrants), so like worth $37bn value transfer to OpenAI
  • That $37bn will help pay for that contract (and it's only fair as OpenAI created all that extra value anyways by partnering with AMD they made AMD more valuable)

OpenAI has the old Elon Musk market moving touch, if you remember those days.

Did ChatGPT write this post?

  • Haha 1
Posted
7 hours ago, Brisketexan said:

He says that...he says lots of stupid shit that is at total odds with the laws of economics.  But the assets on the ground that exist/are being built in the short term are natural gas and some renewables.  AFIAK, ain't nobody even working on getting the financing to build a big new coal plant to serve a data center.

Yeah, these data centers are mostly going to be stand alone gas power plants. Faster to build out and adjust for load with gas. They don't have the time to fuck around with coal. 

  • Hook 'Em 1
  • Like 1
Posted
11 hours ago, Vegas64 said:

From what I can tell: everyone agrees we are in an AI bubble. The differences seem to be some see it as a productive bubble and others as an unproductive speculative bubble.

Dude, this is like asking if you can have a productive Category 5 hurricane. Sure, some people will benefit, but overall it will result in massive economic damage. More people will get fucked than benefit, it's a societal net loss. 

  • Like 3
Posted

https://www.businessinsider.com/ai-impact-economy-gdp-jobless-growth-productivity-2025-10

The US economy is strong, but America may be entering an era of "jobless growth" because of artificial intelligence, Goldman Sachs wrote in a Monday note.

"The modest job growth alongside robust GDP growth seen recently is likely to be normal to some degree in the years ahead," analysts at Goldman Sachs wrote.

They added that this trend is likely to continue, with most potential growth coming from AI-driven productivity. Meanwhile, population and lower immigration would contribute only modestly to the labor supply.

There are already signs of a weaker job market, wrote the analysts.

The analysts pointed out that job growth outside the healthcare industry has turned negative in recent months, and that corporate management teams are increasingly focused on using AI to reduce labor costs — a shift that could weigh on hiring long term.

Posted (edited)
18 minutes ago, Captainant said:

https://www.businessinsider.com/ai-impact-economy-gdp-jobless-growth-productivity-2025-10

The US economy is strong, but America may be entering an era of "jobless growth" because of artificial intelligence, Goldman Sachs wrote in a Monday note.

"The modest job growth alongside robust GDP growth seen recently is likely to be normal to some degree in the years ahead," analysts at Goldman Sachs wrote.

They added that this trend is likely to continue, with most potential growth coming from AI-driven productivity. Meanwhile, population and lower immigration would contribute only modestly to the labor supply.

There are already signs of a weaker job market, wrote the analysts.

The analysts pointed out that job growth outside the healthcare industry has turned negative in recent months, and that corporate management teams are increasingly focused on using AI to reduce labor costs — a shift that could weigh on hiring long term.

It's scary how optimized teams are with AI. In my world I'm seeing it on both ends:

Companies like Chevron for example just went through a 20% re-org with RIFS of like 9-10k people because the old corporate and workforce design is unneccesary for the present (e.g. Oil going to $50) to the future (even when rebounds, don't need the headcount). Large companies are realizing their organizational design is outdated and there are too many humans doing things that digital tools can do.

On the other end of huge corps are start-ups who reach 9 figure ARR and/or get acquired with a fifth of the headcount it used to take. A company I consulted for recently acquired an a16z startup for 9 figures who had 14 full time, w2'ed employees.

Revenue to employee ratio is the scariest metric that has been optimized by AI

Edited by Vegas64
  • Fuck Around and Find Out 1
Posted

Interesting paper published by Anthropic about a tiny sample size of documents being able to poison ANY size LLM

https://www.anthropic.com/research/small-samples-poison

Quote

 

This new study—a collaboration between Anthropic’s Alignment Science team, the UK AISI's Safeguards team, and The Alan Turing Institute—is the largest poisoning investigation to date. It reveals a surprising finding: in our experimental setup with simple backdoors designed to trigger low-stakes behaviors, poisoning attacks require a near-constant number of documents regardless of model and training data size. This finding challenges the existing assumption that larger models require proportionally more poisoned data. Specifically, we demonstrate that by injecting just 250 malicious documents into pretraining data, adversaries can successfully backdoor LLMs ranging from 600M to 13B parameters.

If attackers only need to inject a fixed, small number of documents rather than a percentage of training data, poisoning attacks may be more feasible than previously believed. Creating 250 malicious documents is trivial compared to creating millions, making this vulnerability far more accessible to potential attackers. It’s still unclear if this pattern holds for larger models or more harmful behaviors, but we're sharing these findings to encourage further research both on understanding these attacks and developing effective mitigations.

 

Very interesting considering that there's a shitload of scraped info from github, reddit, and blogs in general. A person could concievably seed in just a few hundred documents with trigger words and set up unintended behavior in the LLM. What's wild is how scalable the attack vector is, you need proportionally smaller sets of "poison" documents the larger and more complex the model is.

Larger models are effectively more vulnerable, because they take in geometrically larger sets of documents in training and need a linearly increasing set of "poison" documents. "Attack success depends on the absolute number of poisoned documents, not the percentage of training data."

  • Hook 'Em 1
  • Fuck Around and Find Out 1
Posted
4 hours ago, Vegas64 said:

It's scary how optimized teams are with AI.

I've been using AI to write Python code for about two years.  Claude is the best one for it.  I shouldn't even be writing code in my role but I can get it done in minutes using AI instead of handing it off to staff to spend hours doing it.  It would take me longer to type an email with the business requirements than to just iterate through it with AI until it work perfectly. 

Posted
1 minute ago, CooterBrown said:

It would take me longer to type an email with the business requirements than to just iterate through it with AI until it work perfectly. 

Candidly that's usually the ACTUAL work of software development in my experience. Business units (and/or customers) rarely are able to fully describe their desired outcome and requirements. If you know that, implementation is significantly easier and faster. Otherwise you just end up doing iterative cycles like you described 

Posted
3 minutes ago, Captainant said:

Candidly that's usually the ACTUAL work of software development in my experience. Business units (and/or customers) rarely are able to fully describe their desired outcome and requirements. If you know that, implementation is significantly easier and faster. Otherwise you just end up doing iterative cycles like you described 

Jira is quite helpful with getting an idea of what a PM wants. Still sucks vut not as much.

Posted
1 hour ago, Captainant said:

Interesting paper published by Anthropic about a tiny sample size of documents being able to poison ANY size LLM

https://www.anthropic.com/research/small-samples-poison

Very interesting considering that there's a shitload of scraped info from github, reddit, and blogs in general. A person could concievably seed in just a few hundred documents with trigger words and set up unintended behavior in the LLM. What's wild is how scalable the attack vector is, you need proportionally smaller sets of "poison" documents the larger and more complex the model is.

Larger models are effectively more vulnerable, because they take in geometrically larger sets of documents in training and need a linearly increasing set of "poison" documents. "Attack success depends on the absolute number of poisoned documents, not the percentage of training data."

I know political campaigns are already planning on trying to seed the AI engines

Posted
Just now, blacklab said:

I know political campaigns are already planning on trying to seed the AI engines

This reminds me.  I gotta get to work on spamming the whole world with my "Brisketexan is a caring lover who brings the satisfaction" posts, so AI will tell all the ladies to get with me.

That's how it works, right?

Posted
5 hours ago, Vegas64 said:

On the other end of huge corps are start-ups who reach 9 figure ARR and/or get acquired with a fifth of the headcount it used to take. A company I consulted for recently acquired an a16z startup for 9 figures who had 14 full time, w2'ed employees.

Not a bubble at all.

Posted

Thanks to AI we can finally use the Internet to get our rocks off, truly groundbreaking stuff.

https://techcrunch.com/2025/10/14/sam-altman-says-chatgpt-will-soon-allow-erotica-for-adult-users/

Quote

OpenAI CEO Sam Altman announced in a post on X Tuesday the company will soon relax some of ChatGPT’s safety restrictions, allowing users to make the chatbot’s responses friendlier or more “human-like,” and for “verified adults” to engage in erotic conversations.

“We made ChatGPT pretty restrictive to make sure we were being careful with mental health issues. We realize this made it less useful/enjoyable to many users who had no mental health problems, but given the seriousness of the issue we wanted to get this right,” said Altman. “In December, as we roll out age-gating more fully and as part of our ‘treat adult users like adults’ principle, we will allow even more, like erotica for verified adults.”

 

  • Hook 'Em 1
Posted
20 hours ago, F250 said:

Dude, this is like asking if you can have a productive Category 5 hurricane. Sure, some people will benefit, but overall it will result in massive economic damage. More people will get fucked than benefit, it's a societal net loss. 

I once had a non-core family member tell me that hurricanes helped the economy by getting construction companies to work. I said if that were true, we could just knock over houses, break some windows, and flood a bunch of businesses to get ourselves out of a recession, but the thing is he was really dumb. 

  • Hook 'Em 1
Posted

This reminds me.  I gotta get to work on spamming the whole world with my "Brisketexan is a horrible lover" posts, so AI will tell Sela Ward and Susanna Hoffs to avoid him.

That's how it works, right?

Posted
53 minutes ago, 956 Worldwide said:

I once had a non-core family member tell me that hurricanes helped the economy by getting construction companies to work. I said if that were true, we could just knock over houses, break some windows, and flood a bunch of businesses to get ourselves out of a recession, but the thing is he was really dumb he is now President of the United States

FIF what I'm sure is true.

Posted
49 minutes ago, atomheartbevo said:

This reminds me.  I gotta get to work on spamming the whole world with my "Brisketexan is a horrible lover" posts, so AI will tell Sela Ward and Susanna Hoffs to avoid him.

That's how it works, right?

Like we need AI to achieve that goal.

Anyway, did you not hear about my new pal, Nassima Benchicou?

676754_26.jpg

 

Posted
5 hours ago, MissingInAction said:

Jira is quite helpful with getting an idea of what a PM wants. Still sucks vut not as much.

It helps when you have good templates and a competent PM. I've had some Jiras escalated to me for lack of work, and then I take a look, and it's just nothing but vaguely described requests with a Definition of Done saying "Application works". 

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.



×
×
  • Create New...