AI vs. software companies

[Spooky]

The best model I've used is the new Google Gemini 3 model but I haven't tried the newest Anthropic 4.6.

[Spooky]

42 minutes ago, gfp said:

 

... that occurred 10 YEARS AGO

 

 

 

Crazy to think about where we will be in another 10 years.

[gfp]

They all have their pros and cons and they tend to leapfrog each other.  For me, today, Gemini 3 Pro is very slow on the server side.  Reminds me of the old days of dial up internet, doing something else while you wait for a page to load.

 

Claude Pro with Opus 4.6 is very good, $200 per year plus tax.  But to get the most out of cowork and associated tools you need to feel comfortable installing an AI on your desktop and even granting it some folder permissions - a lot of people will not feel comfortable with that.  I also found that it is very easy to exhaust your token allowance if you assign it token intensive tasks.  Then you have to wait a few hours to complete the task which is annoying if you paid $200.

 

ChatGPT's newest model is a big improvement.  They excel sometimes when Gemini struggles.

 

I don't use Grok as often but have heard it is improving very quickly and is very good.

 

Those are the big 4.  Experiment with the free versions and see which ones you prefer.  There is a big learning curve involved in getting the most out of each.

 

The Gemini 3 Pro ecosystem also integrates with google drive and Notebook LM if that is a consideration

[Eng12345]

Ive found free ChatGPT to be more useful for daily conversational general use type activities and Claude better for more technical work. 

 

Been thinking about pulling the trigger on Claude Pro. 

Edited February 13 by Eng12345

[frommi]

1 hour ago, Spooky said:

 

I'm not so sure about this. Look at Google's Alpha Go - it came up with a totally new strategy to win a game that has been played by humans for over 2,000 years.

Games like Chess or Go differ from real life because they have a clear set of rules that define when a game is won or lost. This structure can be used for training purposes or as a reward function.

[Spooky]

14 minutes ago, frommi said:

Games like Chess or Go differ from real life because they have a clear set of rules that define when a game is won or lost. This structure can be used for training purposes or as a reward function.

 

I don't know, for a long time Go was thought to be too complex of a game for AI to master. Go has more possible positions than atoms in the observable universe, making it impossible for traditional "brute force" AI to win by calculating every move.

 

Hey Siri, maximize shareholder value!

Edited February 13 by Spooky

[frommi]

Alpha Go used the end result of a game as a reward, it was either win or loss. And than it trained a neural net on that outcome and played millions of games against itself and learned the game that way. 
 

[gfp]

What in the world is the use of debating problems that Ai solved over a decade ago as a way to understand today or next month or two years from now?

[Castanza]

28 minutes ago, gfp said:

What in the world is the use of debating problems that Ai solved over a decade ago as a way to understand today or next month or two years from now?

Has there been any significant “move 37” like occurrences since? I was reading through the AlphaEvolve one roughly 8 months ago and it’s more of the algo making an improvement with already known algorithms.so quite good at optimizing workflows but not creating or deriving new solutions. So far it seems the upper limits are getting to solutions with known algorithms at much faster speeds than humans (think 1000 years of effort). So extremely important and a huge breakthrough for sure. But not true AGI. 
 

That’s just what I’m aware of, very possibly more I haven’t read up on. What are you looking at?

Edited February 13 by Castanza

[gfp]

I mean, they gave Demmis a freaking nobel prize for the protein folding problem so that would be the next ancient example that comes to mind...

[Castanza]

14 minutes ago, gfp said:

I mean, they gave Demmis a freaking nobel prize for the protein folding problem so that would be the next ancient example that comes to mind...

 

A perfect problem for AI to solve no doubt; but it's a narrowly focused tool took a master class team and the worlds best AI lab to build. It is still a "brute-force" like model that utilizes additional data structures and approaches to basically pull forward prediction models to be as accurate as possible BEFORE a human has to interpret them. For all intended purposes this shows the best use cases and limitations of current AI models and doesn't move us any close to AGI. 

 

But AGI is not needed to change the world

[gfp]

1 minute ago, Castanza said:

 

A perfect problem for AI to solve no doubt; but it's a narrowly focused tool took a master class team and the worlds best AI lab to build. It is still a "brute-force" like model that utilizes additional data structures and approaches to basically pull forward prediction models to be as accurate as possible BEFORE a human has to interpret them. For all intended purposes this shows the best use cases and limitations of current AI models and doesn't move us any close to AGI. 

 

But AGI is not needed to change the world

 

Half a decade ago!!!

 

 

[MungerWunger]

Nice discussion here between @joinyellowbrick and @ArenaManCapital:

https://x.com/joinyellowbrick/status/2022055862206882011?s=20

 

Edited February 13 by MungerWunger

[Spekulatius]

3 hours ago, gfp said:

 

Half a decade ago!!!

 

 

I am not sure what you are implying. A breakthrough was made a decade ago for a particular problem and then we need another breakthrough to go from there which took another decade.

 

Or maybe the current LLM’s  instead of maxing out at some point will become a runaway train that improves itself to get beyond what humans can ever accomplish?

[Castanza]

6 hours ago, gfp said:

 

Half a decade ago!!!

 

 

I’m not following you here. It was your example lol Perhaps we are not understanding each others points? 
 

 

Edited February 14 by Castanza

[MungerWunger]

https://fortune.com/2026/02/13/when-will-ai-kill-white-collar-office-jobs-18-months-microsoft-mustafa-suleyman/ 

"Microsoft AI chief gives it 18 months—for all white-collar work to be automated by AI"

 

https://www.cnbc.com/2026/02/06/google-microsoft-pay-creators-500000-and-more-to-promote-ai.html

"Companies including Microsoft and Google have paid creators between $400,000 and $600,000 for long-term partnerships spanning several months, CNBC has learned."

 

Why do they need to promote AI so heavily if white collar jobs are gone in just 1.5 years from now?

[Spekulatius]

1 hour ago, MungerWunger said:

"Microsoft AI chief gives it 18 months—for all white-collar work to be automated by AI"

I would like to take the other side of this bet.  
 

Also, if I were interviewer, I would have asked the him question how Microsoft HR is dealing with the problem. Are they starting the layoffs now or do they intend to do it all at once in 18 months? Is he going to be the last one to turn off the lights?

[frommi]

3 hours ago, MungerWunger said:

Why do they need to promote AI so heavily if white collar jobs are gone in just 1.5 years from now?

Because they are in the business of raising money. Listen to the CEO from Anthropic (interview in the other threat) what he tells us about revenue and training cost is eye opening. Each model each year needs more money for training and they are not able to generate enough revenue to even cover the cost of the current model. At some point they will ran out of money. Thats capital destruction at its finest.

[Spekulatius]

8 hours ago, frommi said:

Because they are in the business of raising money. Listen to the CEO from Anthropic (interview in the other threat) what he tells us about revenue and training cost is eye opening. Each model each year needs more money for training and they are not able to generate enough revenue to even cover the cost of the current model. At some point they will ran out of money. Thats capital destruction at its finest.

Well, he works for Microsoft and they do not need raise money.

 

If these guys are remotely right, we need to buy puts on pretty much everything, especially financials. When half the people go in default, pretty much any financial stocks should crash and would be a zero.

 

Cash would be relatively safe in such a deflationary scenario, at least at first.

[frommi]

Msft is now part of that ecosystem, they benefit a lot from the cloud bookings of OpenAI and Anthropic. And at the same time they finance their own revenue by financing these AI companies. I really hope that they are not right. 
Its really funny that everyone hypes them up, but doesnt think about the effects that has on everyone on the planet. Yeah maybe 2026 is the year of the Put

Edited February 14 by frommi

[UK]

21 hours ago, gfp said:

a lot of people will not feel comfortable with that

This is a good point. I personally came to the conclusion, that I can only share my data with Google, because it has mostly all my data anyway:)

[Spekulatius]

11 hours ago, MungerWunger said:

https://fortune.com/2026/02/13/when-will-ai-kill-white-collar-office-jobs-18-months-microsoft-mustafa-suleyman/ 

"Microsoft AI chief gives it 18 months—for all white-collar work to be automated by AI"

 

https://www.cnbc.com/2026/02/06/google-microsoft-pay-creators-500000-and-more-to-promote-ai.html

"Companies including Microsoft and Google have paid creators between $400,000 and $600,000 for long-term partnerships spanning several months, CNBC has learned."

 

Why do they need to promote AI so heavily if white collar jobs are gone in just 1.5 years from now?

I think the answer is obvious. They need the people to believe in this future, in order to finance it. Basic reflexivity at work here.

 

The future will only happen if people believe in it. reflexivity 101.

 

Example from the past, of the telecom boom would have lasted 3 years longer, pretty much everyone in America would have had FTTH in 2007 or 2008.

 

 

 

 

 

 

[frommi]

Its about Msft, but it touches a lot of AI points.

[brobro777]

6 hours ago, frommi said:

Msft is now part of that ecosystem, they benefit a lot from the cloud bookings of OpenAI and Anthropic. And at the same time they finance their own revenue by financing these AI companies. I really hope that they are not right. 
Its really funny that everyone hypes them up, but doesnt think about the effects that has on everyone on the planet. Yeah maybe 2026 is the year of the Put

SPX returns 15% plus for past three years straight, so take a breather for -10% in 2026, While PM marlboro cigs hit record highs? not bad baby! 

 

 

[UK]

3 hours ago, frommi said:

Its about Msft, but it touches a lot of AI points.

Gemini: 

This video by Drew Cohen (former Goldman Sachs and Capital Group analyst) provides a deep dive into Microsoft's current business transformation as it shifts from a traditional software company to an "AI-first" company.
Core Business & Financial Overview
Revenue Streams: Microsoft generated $35 billion in revenue over the last 12 months, with a 47% EBIT margin [06:24].
Segment Risks: The "Productivity and Business Process" segment (Office 365, LinkedIn) holds a $70 billion profit pool that is now at risk as AI agents potentially replace the need for traditional software seats [07:16].
The Valuation Gap: There is a significant discrepancy between its trailing earnings multiple (25x) and its free cash flow multiple (46x), largely due to massive capital expenditures (CapEx) for AI infrastructure [01:52].
The AI Transformation & Competitive Risks
Software "Collapse": CEO Satya Nadella has suggested that software apps may "collapse" because AI agents can create their own logic and bypass traditional software interfaces [21:20].
Platform as a Service (PaaS) Threat: Historically, Microsoft’s cloud (Azure) has been "sticky" because developers build on its platform layer. AI agents can now circumvent this layer to go directly to databases, potentially reducing this competitive advantage [22:00].
The "Existential" Risk: The primary threat is if a single competitor achieves "godlike AI" that is so superior it renders Microsoft’s distribution and existing tools irrelevant [29:41].
Infrastructure & Accounting Realities
The "Fiber Optic" Parallel: Microsoft is spending ~$150 billion on data centers. There is a risk of overcapacity—similar to the fiber optic bubble of the 90s—where massive investments are made before demand fully arrives [39:30].
Earnings Headwind: Due to accounting rules, the depreciation of these massive hardware investments will eventually catch up to the income statement. This will mathematically act as a drag on earnings, regardless of revenue growth [43:36].
Five Key Takeaways for Investors [52:24]:
Cannibalization: AI agents are replacing human "seats" in the software business.
PaaS Vulnerability: The stickiness of Azure's platform layer is under threat.
Overcapacity Risk: The risk of overshooting demand for data centers.
Accounting Headwinds: Depreciation will inevitably impact future earnings.
Commoditization: Microsoft’s success depends on AI models becoming a commodity so they can win through their massive distribution network.