Artificial Intelligence

[WayWardCloud]

Thanks for clarifying!

I agree that "AGI" whatever that means is a very over-hyped and vague concept.

 

I don't know, however, what "real" intelligence means.

Every major scientific revolution I can think of has had the effect of humbling us.

Heliocentrism (Earth isn't the center of the universe)

Evolution (Man is just another animal)

The subconscious (humbling of the conscious mind / the ego / the illusion of rationality)

 

Maybe the next big humbling is that our precious intelligence is some type of LLM that can be done just the same in a lab and actually improved upon by building brains that are much bigger than what the cranium can sustain. When you hear a child speak they start with stringing together random sounds and then words they don't fully "understand". As they grow they just get better and better at making mostly structurally sound sentences. Maybe we just forgot how we all started "thinking" when we were kids. Even with fully developed brains, most adults I hear give their "opinion" just regurgitate and sum up (poorly) the last 2-3 things they've heard/read and they call that thinking and they believe it's super unique and precious...

 

Anyway, that was a digression. The main point I want to get across is I would suggest focusing on what the new tool (AI) does do and not on what it doesn't do. The current capabilities are already able to create a lot of value once they are fully integrated into the production chain and understood by the workers and those capabilities are growing every day. Maybe some things do stay out of its reach and only humans can do them, that can be true and also that trillions in labor can be automated / augmented both at the same time. This reminds me of discussions here a couple years ago about self driving when some of the forumers arguments were that it would never happen because of some crazy edge cases they could think of. It can be true that only humans can drive in a blizzard or an icy road or when sun flares hit the car at the worst angle or whatever and at the same time true that most boring city commuting can be automated. We are seeing it now with Waymo/Tesla Robotaxi.

 

So even if LLMs keep fumbling things that are very easy to us forever (I doubt that's the case but who knows) they also already are incredibly better than us at other tasks and their capabilities are growing so I find it hard to believe they won't create a tremendous amount of value.

Edited September 6, 2025 by WayWardCloud

[Eldad]

2 minutes ago, WayWardCloud said:

Thanks for clarifying!

I agree that "AGI" whatever that means is a very over-hyped and vague concept.

 

I don't know, however, what "real" intelligence means.

Every major scientific revolution I can think of has had the effect of humbling us.

Heliocentrism (Earth isn't the center of the universe)

Evolution (Man is just another animal)

The subconscious (humbling of the conscious mind / the ego)

 

Maybe the next big humbling is that our precious intelligence is some type of LLM that can be done just the same in a lab and actually improved upon by building brains that are much bigger than what the cranium can sustain. When you hear a child speak they start with stringing together random sounds and then words they don't fully "understand". As they grow they just get better and better at making mostly structurally sound sentences. Maybe we just forgot how we all started "thinking" when we were kids. Even with fully developed brains, most adults I hear give their opinion mostly just regurgitate and sum up the last 2-3 things they've heard/read and they call that thinking and they believe it's super unique and precious...

 

Anyway, that was a digression. The main point I want to get across is I would suggest focusing on what the new tool (AI) does do and not on what it doesn't do. The current capabilities are already able to create a lot of value once they are fully integrated into the production chain and understood by the workers and those capabilities are growing every day. Maybe some things do stay out of its reach and only humans can do them, that can be true and also that trillions in labor can be automated / augmented both at the same time.

 

This reminds me of discussions here a couple years ago about self driving when some of the forumers arguments were that it would never happen because of some crazy edge cases they could think of. It can be true that only humans can drive in a blizzard or an icy road or when sun flares hit the car at the worst angle or whatever and at the same time true that most boring city commuting can be automated. We are seeing it now with Waymo/Tesla Robotaxi.

I am really not sure of the use case honestly. There is certainly value in being able to process lots of data and apply even some very elementary logic to that data, but if it gives you a hallucination even 10% of the time it is essentially worthless is it not? 

 

Even if I am wrong, these basic things that it cannot do correctly when weighted against the time and funding cost and crazy pie in the sky promises just seems so out of balance from an investment prospective. 

 

I also do not think all scientific progress has degraded man, but we live in maybe the most materialistic age ever so that is a common notion and even a hope I think for some reason. But that is another discussion. 

[WayWardCloud]

2 minutes ago, Eldad said:

if it gives you a hallucination even 10% of the time it is essentially worthless is it not? 

 

When you ask a human coworker to do a task how often do they make mistakes on the assignment or confidently answer something completely wrong? For me it's hard to find someone that will do every task I ask for. I'll send an email asking five questions and get the answer for maybe two of those back and they feel like they've completed the task. It's a lot of trial and error and double checking other people's work and asking for the same thing over and over again. Sometimes they even mess up in the exact way I predicted and specifically asked not to mess up. We have chains of command with lots of redundancy and double checking for that reason. There are tasks where a 10% error rate is totally fine (sorting trash) and tasks where it isn't (surgery). Actually the Pareto principle states that you are most productive when you stop once about 80% of the goal is reached and move to the next one. Maybe tomorrow's hot job is "hallucination hunter" where OCD/perfectionist people are in super high demand, that'd be really good for me 

 

2 minutes ago, Eldad said:

I also do not think all scientific progress has degraded man

 

"humbling" is a good thing in my vocabulary  I do not think scientific progress has degraded man in any way. I think we tend to romanticize early humanity a lot and that it's been mostly a painful and brutal shit show that we are slowly improving upon thanks to the scientific method.

[Eldad]

15 minutes ago, WayWardCloud said:

 

When you ask a human coworker to do a task how often do they make mistakes on the assignment or confidently answer something completely wrong? For me it's hard to find someone that will do every task I ask for. I'll send an email asking five questions and get the answer for maybe two of those back and they feel like they've completed the task. It's a lot of trial and error and double checking other people's work and asking for the same thing over and over again. Sometimes they even mess up in the exact way I predicted and specifically asked not to mess up. We have chains of command with lots of redundancy and double checking for that reason. There are tasks where a 10% error rate is totally fine (sorting trash) and tasks where it isn't (surgery). Actually the Pareto principle states that you are most productive when you stop once about 80% of the goal is reached and move to the next one. Maybe tomorrow's hot job is "hallucination hunter" where OCD/perfectionist people are in super high demand, that'd be really good for me 

 

 

"humbling" is a good thing in my vocabulary  I do not think scientific progress has degraded man in any way. I think we tend to romanticize early humanity a lot and that it's been mostly a painful and brutal shit show that we are slowly improving upon thanks to the scientific method.

Good points. I am an OCD perfectionist as well and just really hate giving someone a wrong answer or even an educated guess, but we are a minority it seems. 
 

People are dumb and lazy now and can’t even bother to read their boss’s instructions carefully so this will be a good substitute doesn’t seem like the best path for civilization LOL. 

Edited September 6, 2025 by Eldad

[treasurehunt]

1 hour ago, Eldad said:

I am really not sure of the use case honestly. There is certainly value in being able to process lots of data and apply even some very elementary logic to that data, but if it gives you a hallucination even 10% of the time it is essentially worthless is it not? 

 

I'd say this is far from worthless; in fact, it's probably very valuable. Typically, it is much easier to check if an answer makes sense than it is to come up with an answer in the first place.

[Milu]

2 hours ago, treasurehunt said:

 

I'd say this is far from worthless; in fact, it's probably very valuable. Typically, it is much easier to check if an answer makes sense than it is to come up with an answer in the first place.

Yes I like to treat it as a junior employee, get it to do the grunt work and come up with an answer, but then be sure to check it makes sense.

[Dalal.Holdings]

It’s okay guys, I’m sure this will be all worth it…and if OpenAI spends it, everyone else has to in order to keep up in the “race” !

 

 

 

[Dalal.Holdings]

12 hours ago, vinod1 said:

So why mix what you know with a high degree of certanity - current Mag 7 profits with something we know vaguely and imprecisely - GDP calcuations?

 

 My point being, don't use these percent of GDP numbers to influence your investment decisions. 

 

Vinod

 

Show me the Mag 7 profits being earned from AI… this is not something anyone “knows with a high degree of certainty”

 

We can clearly see OpenAI is burning through lots of cash, so I expect the same with AI ventures of the Mag 7 (excluding companies like Apple which are not building a model)

 

Edited September 6, 2025 by Dalal.Holdings

[Eldad]

4 hours ago, Dalal.Holdings said:

 

Show me the Mag 7 profits being earned from AI… this is not something anyone “knows with a high degree of certainty”

 

We can clearly see OpenAI is burning through lots of cash, so I expect the same with AI ventures of the Mag 7 (excluding companies like Apple which are not building a model)

 

Mainly agree with you but GOOGL, MSFT, AMZN at least have the cloud service business that is somewhat aligned with all of this build out and can but saas stuff on top of it. 
 

META seems to be on tilt and is kind of worrying me. Open AI is probably not long for this world. 

[Dalal.Holdings]

I surmise OpenAI is probably doing the best in terms of returns on AI as they are by far the most well known LLM and models and people subscribe to them. I'm not sure about the others. The cloud businesses might have been fine without AI or by contracting third party AI...

 

AMZN's capex is of major concern as their FCF ttm has dwindled to just $18B (their capex is north of $100B) ...for a $2.4 Trillion market cap... and if you subtract stock based comp, FCF is negative...

 

Edited September 6, 2025 by Dalal.Holdings

[Dalal.Holdings]

LOL the numbers keep getting bigger…

 

If you’re a META investor, you have to hope he’s just being hyperbolic for Trump. Otherwise say adios to $600B (or more)…by 2028??? LMAO 

 

 

Edited September 6, 2025 by Dalal.Holdings

[LC]

I think google or microsoft will be the big AI winners - transitioning enterprise clients to a full solution suite: cloud, LLMs, and the surrounding services for data, privacy, & cybersecurity. The real cherry on top is access to enterprise data to train their models.  If they can manage to anonymize and encrypt that data and convince enterprises to allow them training access (or just do it anyways), I think that is what propels this technology to long-term commercialization. 

[Dalal.Holdings]

It’s important to remember that in the year 2000 when the internet bubble was peaking, everyone knew the internet would change the world, but Google was a PhD thesis no one knew about and Facebook/Instagram/TikTok/YouTube/so many others were not even born yet.

 

Instead, everyone in the year 2000 though the big internet winners would be CSCO and AOL. I guess you can add AMZN as the only one that really made it through that period, but you had to suffer a big drawdown.

 

My point being that those who think they can predict who the big winner of AI will be may be very badly wrong…

[vinod1]

 

 

On 9/6/2025 at 9:12 AM, Dalal.Holdings said:

 

Show me the Mag 7 profits being earned from AI… this is not something anyone “knows with a high degree of certainty”

 

We can clearly see OpenAI is burning through lots of cash, so I expect the same with AI ventures of the Mag 7 (excluding companies like Apple which are not building a model)

 

 

The Mag 7 reported earnings of $550 billion in 2024 and paid taxes. It is as high a degree of certanity as you can get.

 

A $ is a $ whether it is earned from AI or something else. Their track record has been quite good.

 

Now, we have a technology that might have a massive impact on everything from jobs to companies and all the way up to nations themselves. There are only a few, I suspect fewer than a dozen entities in the world that can spend the amount of money these companies can spend on this technology. 

 

If these companies spend $2 trillion to have a shot at this technology and they waste half of it, it is still only 2 or 3 years of earnings that would be lost. Most of their investments can be used for their core operations anyway, so it would never be a complete loss.

 

I am glad these companies are making these investments.

 

Vinod

 

  

 

 

[vinod1]

23 hours ago, Dalal.Holdings said:

It’s important to remember that in the year 2000 when the internet bubble was peaking, everyone knew the internet would change the world, but Google was a PhD thesis no one knew about and Facebook/Instagram/TikTok/YouTube/so many others were not even born yet.

 

Instead, everyone in the year 2000 though the big internet winners would be CSCO and AOL. I guess you can add AMZN as the only one that really made it through that period, but you had to suffer a big drawdown.

 

My point being that those who think they can predict who the big winner of AI will be may be very badly wrong…

 

You can also make the argument that many took the wrong lessons, got traumatized by the experience, swore off from making this mistake ever again and underperformed for the next 2 decades. 

 

You might actually making a point in support of massive investments by these companies! They are aware of this history, constantly worry about it, and jump at the earliest glimpse of some new technology that could disrupt their position.

 

 

Edited September 8, 2025 by vinod1

[Dalal.Holdings]

1 hour ago, vinod1 said:

 

 

 

The Mag 7 reported earnings of $550 billion in 2024 and paid taxes. It is as high a degree of certanity as you can get.

 

A $ is a $ whether it is earned from AI or something else. Their track record has been quite good.

 

Now, we have a technology that might have a massive impact on everything from jobs to companies and all the way up to nations themselves. There are only a few, I suspect fewer than a dozen entities in the world that can spend the amount of money these companies can spend on this technology. 

 

If these companies spend $2 trillion to have a shot at this technology and they waste half of it, it is still only 2 or 3 years of earnings that would be lost. Most of their investments can be used for their core operations anyway, so it would never be a complete loss.

 

I am glad these companies are making these investments.

 

Vinod

 

  

 

 

 

What is the multiple you are paying on Mag 7 earnings? 

 

Are the earnings in real cash (free cash flow less stock based comp) or do they include assets that could be written down in the future if profits do not materialize?

 

What will the returns on capital of these businesses be going forward? Is it wise to reinvest at the rate that they are at those returns on capital ?

 

Some things to ponder.

[treasurehunt]

3 hours ago, Dalal.Holdings said:

 

What is the multiple you are paying on Mag 7 earnings? 

 

Are the earnings in real cash (free cash flow less stock based comp) or do they include assets that could be written down in the future if profits do not materialize?

 

What will the returns on capital of these businesses be going forward? Is it wise to reinvest at the rate that they are at those returns on capital ?

 

Some things to ponder.

 

All interesting questions.

 

ChatGPT gave me the following numbers for the Mag 7. I verified some independently, so they are probably in the ballpark.

 

Market cap: $17.5T

GAAP Net Income: $551B

FCF: $380B

FCF - SBC: $288B

 

Using these three measures of earnings, the P/E ratios work out to 32, 46 and 61. It seems to me that the market is expecting AI investments to work out well for these companies.

[Eldad]

8 hours ago, treasurehunt said:

 

All interesting questions.

 

ChatGPT gave me the following numbers for the Mag 7. I verified some independently, so they are probably in the ballpark.

 

Market cap: $17.5T

GAAP Net Income: $551B

FCF: $380B

FCF - SBC: $288B

 

Using these three measures of earnings, the P/E ratios work out to 32, 46 and 61. It seems to me that the market is expecting AI investments to work out well for these companies.

Market CAP for Mag 7 is easiest to verify. 
As of this morning I get 19.83T

 

Just off 2.33 Trillion. LOL. It is a BS engine. 

[Longnose]

On 9/6/2025 at 6:27 PM, Dalal.Holdings said:

It’s important to remember that in the year 2000 when the internet bubble was peaking, everyone knew the internet would change the world, but Google was a PhD thesis no one knew about and Facebook/Instagram/TikTok/YouTube/so many others were not even born yet.

 

Instead, everyone in the year 2000 though the big internet winners would be CSCO and AOL. I guess you can add AMZN as the only one that really made it through that period, but you had to suffer a big drawdown.

 

My point being that those who think they can predict who the big winner of AI will be may be very badly wrong…

I think the better parallel for Google, Microsoft, and the other tech giants isn't to the internet companies of 2000, but to the railroad companies.

The infrastructure required for modern AI, the vast data centers, specialized chips, and global networks is so immense and costly that it's more like laying railroad tracks across the country than building a new website. Companies like Google and Microsoft have been building this infrastructure for years. They have the "rails" already in place, giving them a significant advantage.

existing cloud services (Google Cloud, Microsoft Azure, Amazon AWS) are the modern equivalent of the railroad lines. They are the essential infrastructure that enables all new AI applications, and is extremely difficult for new competitors to build that kind of scale and distribution.

This isn't to say a new player can't build a great "train" (a killer AI application), but they'll likely have to run it on someone else's tracks. The real power, and the "moat," lies with the companies that own the underlying infrastructure and distribution. In that sense, the Googles and Microsofts are more like the Union Pacific and Central Pacific of the AI era they're the enablers who will likely profit regardless of which AI applications become the biggest success.

[Eldad]

4 minutes ago, Longnose said:

I think the better parallel for Google, Microsoft, and the other tech giants isn't to the internet companies of 2000, but to the railroad companies.

The infrastructure required for modern AI, the vast data centers, specialized chips, and global networks is so immense and costly that it's more like laying railroad tracks across the country than building a new website. Companies like Google and Microsoft have been building this infrastructure for years. They have the "rails" already in place, giving them a significant advantage.

existing cloud services (Google Cloud, Microsoft Azure, Amazon AWS) are the modern equivalent of the railroad lines. They are the essential infrastructure that enables all new AI applications, and is extremely difficult for new competitors to build that kind of scale and distribution.

This isn't to say a new player can't build a great "train" (a killer AI application), but they'll likely have to run it on someone else's tracks. The real power, and the "moat," lies with the companies that own the underlying infrastructure and distribution. In that sense, the Googles and Microsofts are more like the Union Pacific and Central Pacific of the AI era they're the enablers who will likely profit regardless of which AI applications become the biggest success.

Except it’s digital and not physical and can become redundant or obsolete very quickly. The non AI enabled data centers built just a few years ago are already becoming obsolete. I get what you are saying but it may be an over simplification. 

[Eldad]

13 minutes ago, Eldad said:

Except it’s digital and not physical and can become redundant or obsolete very quickly. The non AI enabled data centers built just a few years ago are already becoming obsolete. I get what you are saying but it may be an over simplification. 

Perfect example is the 2000 bubble with the fiber buildout. Then CSCO switches made most of it worthless. 

[Longnose]

15 minutes ago, Eldad said:

Except it’s digital and not physical and can become redundant or obsolete very quickly. The non AI enabled data centers built just a few years ago are already becoming obsolete. I get what you are saying but it may be an over simplification. 

Maybe... But im happy to continue to bet that MSFT and GOOG will own all those rails and buy all of NVDA's inventory and even if its constantly evolving and going obsolete. Those rails / data are as real a moat as any. 

[treasurehunt]

1 hour ago, Eldad said:

Market CAP for Mag 7 is easiest to verify. 
As of this morning I get 19.83T

 

Just off 2.33 Trillion. LOL. It is a BS engine. 

 

Yikes! I took another look at its work. Market caps for NVDA, AMZN, TSLA and META are accurate, but ChatGPT is way off on AAPL, MSFT and GOOG. Didn't expect that.

 

I just tried Gemini as well. ChatGPT actually does better than Gemini, which thinks the total market cap is $16.57 trillion. 

[Dalal.Holdings]

1 hour ago, Longnose said:

I think the better parallel for Google, Microsoft, and the other tech giants isn't to the internet companies of 2000, but to the railroad companies.

The infrastructure required for modern AI, the vast data centers, specialized chips, and global networks is so immense and costly that it's more like laying railroad tracks across the country than building a new website. Companies like Google and Microsoft have been building this infrastructure for years. They have the "rails" already in place, giving them a significant advantage.

existing cloud services (Google Cloud, Microsoft Azure, Amazon AWS) are the modern equivalent of the railroad lines. They are the essential infrastructure that enables all new AI applications, and is extremely difficult for new competitors to build that kind of scale and distribution.

This isn't to say a new player can't build a great "train" (a killer AI application), but they'll likely have to run it on someone else's tracks. The real power, and the "moat," lies with the companies that own the underlying infrastructure and distribution. In that sense, the Googles and Microsofts are more like the Union Pacific and Central Pacific of the AI era they're the enablers who will likely profit regardless of which AI applications become the biggest success.

When Cornelius Vanderbilt owned the NY central railroad, he pretty much had the only major railroad going into Manhattan Grand Central Terminal…

 

UNP had access to the west.

 

With AI, you have a bunch of railroads that take you to about the same place—as we see when you ask them “what is the market cap of the mag 7?” They all bring you to nearly the same place.

 

Furthermore, the “rails” or chips from NVDA seem to depreciate rapidly as new chips are released which means once you have the rails down, you’ll need to spend lots of capex to replace them again otherwise you will no longer be competitive …

 

IMO Apple is wise to sit this race out

[Dalal.Holdings]

1 hour ago, Longnose said:

Maybe... But im happy to continue to bet that MSFT and GOOG will own all those rails and buy all of NVDA's inventory and even if its constantly evolving and going obsolete. Those rails / data are as real a moat as any. 

 

One of the big reasons why Buffett bought BNSF is because "the rails are already put down" and don't need to be replaced. It's also much harder to build new railways because of Nimbyism/property ownership/etc.

 

These things are not true for your AI:railroad analogy: the rails need to be replaced frequently ($$$ recurring capex/rapid depreciation), and it's not nearly as hard for a new entrant to come in and buy the latest NVDA chips (or maybe not: Deepseek) and replicate the LLMs and achieve much of the same things (build their own railroad)

 

Also, as I noted, these LLM's are like a bunch of railroad tracks that are duplicative and run the same routes next to each other...

 

Finally, Railroads were often terrible investments in the 1800s and much of the 20th century as well (lots of investors lost lots of capital)--despite railroads being game changers that advanced the world significantly...there is a parallel with airlines in that respect too

 

 

Edited September 8, 2025 by Dalal.Holdings