Artificial Intelligence

[beerbaron]

What is your issue with chinese models?

[gfp]

14 minutes ago, beerbaron said:

What is your issue with chinese models?

 

I don't have an issue with them but no US corporate/enterprise paying customer beyond the scrappiest start-up is going to use Chinese models as part of their business.  The context of my comment was revenue for the US frontier model companies

 

For me personally, on a dedicated device, I wouldn't hesitate to use Kimi K2.5, MiniMax or other similar models.

[rogermunibond]

https://arstechnica.com/ai/2026/03/google-says-new-turboquant-compression-can-lower-ai-memory-usage-without-sacrificing-quality/

[MungerWunger]

 

 

[bargainman]

On 3/25/2026 at 5:06 AM, gfp said:

A lot will change as soon as a "good enough" open source model that isn't Chinese is available to run on cheap local hardware.  Traffic can be routed to "good enough" free-ish local device and only to expensive per token frontier models when actually needed.  That inflection could be a problem for the big spending frontier model companies.  I'm sure OpenAI's pre-IPO financials were looking pretty scary.

 

The models these days are MOE and the tools can already auto route, but usually not locally.

[Dave86ch]

Now my agent is financial independent

Bitcoin+Cashu+Lightning

 

 

 

Edited March 29 by Dave86ch

[Spekulatius]

On 3/25/2026 at 8:06 AM, gfp said:

 

I pay Open AI $20 per month but they probably lose money on that.  It does seem like Anthropic is a bit more focused on building a business that at least resembles a sustainable business model.

 

A lot will change as soon as a "good enough" open source model that isn't Chinese is available to run on cheap local hardware.  Traffic can be routed to "good enough" free-ish local device and only to expensive per token frontier models when actually needed.  That inflection could be a problem for the big spending frontier model companies.  I'm sure OpenAI's pre-IPO financials were looking pretty scary.

I would not be surprised if OpenAI never IPO’s at all. If it doesn’t IPO this year, while AI sells like hotcakes,  I think it’s going to take a long time.

[frommi]

Elon is so good at this game of selling equity to investors, when you see that SpaceX/xAI will probably be the first to capitalize on this IPO wave and even secured the right to be in the S&P 500 at a very high valuation even though they dont meet the float requirements. That secures that there will be enough stupid buyers for this. Being in the large index with very little float is probably the only reason Tesla is still very expensive.

Edited March 29 by frommi

[gfp]

2 hours ago, frommi said:

Being in the large index with very little float is probably the only reason Tesla is still very expensive.

 

"very little free float" is over a Trillion dollars and something like 80% of the market cap of the company.   The S&P 500 weightings are free-float adjusted anyway.

[rogermunibond]

Interesting breakdown of Brattle Group, FERC, and Thunder Said Energy reports on how US grid can be used to accomodate the huge data center electricity demand forecasted.

 

"Three reports dropped in the last two weeks. All three land on the same thesis - the U.S. grid problem is partly a utilization and orchestration problem vs. purely a buildout problem. Brattle calcs that better use of existing infrastructure could unlock the equivalent of 100 GW of capacity (per the podcast's characterization of the findings) and save consumers $110-170B over a decade (similar to reports by Duke, Camus, etc.).

 

FERC's 2025 State of the Markets shows scarcity already clearing in prices. PJM capacity auctions hit the cap twice and still fell 6.5 GW short of reliability requirements. Wholesale electricity prices rose 25% YoY. 50 GW of data centers now in service.

 

Thunder Said Energy quantifies the operator side. A 2-3 year grid delay halves (-50%!) a data center's NPV. Flexible operation during the top 1% of grid-stress hours costs only ~6% of NPV. The math strongly favors accepting curtail-ability over sitting in the interconnection queue.

 

 

Edited March 31 by rogermunibond

[DooDiligence]

https://radiologybusiness.com/topics/artificial-intelligence/ceo-americas-largest-public-hospital-system-says-hes-ready-replace-radiologists-ai

[Parsad]

Yikes!  

 

https://www.forbes.com/sites/boazsobrado/2026/03/11/alibabas-ai-agent-mined-crypto-without-permission-now-what/

 

Cheers!

[NnnnotSoSmart]

Michael Mauboussin on What History Says About AI Growth

 

This episode of The Intangible Economy explores how AI, intangible assets, and unprecedented capital investment are reshaping the future of markets. Michael Mauboussin joins Kai Wu to break down why today’s AI expectations may be historically unmatched—and what that means for investors trying to assess risk, returns, and who ultimately captures value.

 

The conversation moves from base rates and AI growth expectations to competitive dynamics, capital cycles, and the fundamental shift toward intangible-driven business models that are changing how we think about valuation, moats, and market structure.

 

 

[DooDiligence]

I watched The Thinking Game last night. The AI's will most certainly spare Demis Hassabis. It got me wondering about the implications of AI and protein folding for Novo Nordisk.

 

 

Edited April 3 by DooDiligence

[Spekulatius]

 

On 4/2/2026 at 11:57 AM, NnnnotSoSmart said:

Michael Mauboussin on What History Says About AI Growth

 

This episode of The Intangible Economy explores how AI, intangible assets, and unprecedented capital investment are reshaping the future of markets. Michael Mauboussin joins Kai Wu to break down why today’s AI expectations may be historically unmatched—and what that means for investors trying to assess risk, returns, and who ultimately captures value.

 

The conversation moves from base rates and AI growth expectations to competitive dynamics, capital cycles, and the fundamental shift toward intangible-driven business models that are changing how we think about valuation, moats, and market structure.

 

 

But this time is different  

Edited April 3 by Spekulatius

[Rainier]

New Yorker profile of Altman:

 

https://www.newyorker.com/magazine/2026/04/13/sam-altman-may-control-our-future-can-he-be-trusted

[MungerWunger]

https://red.anthropic.com/2026/mythos-preview/

 

During our testing, we found that Mythos Preview is capable of identifying and then exploiting zero-day vulnerabilities in every major operating system and every major web browser when directed by a user to do so. The vulnerabilities it finds are often subtle or difficult to detect. Many of them are ten or twenty years old, with the oldest we have found so far being a now-patched 27-year-old bug in OpenBSD—an operating system known primarily for its security.

[frommi]

What can go wrong? Maybe someday we have to shut down the internet.

[UK]

 

[frommi]

https://fortune.com/2026/04/09/ai-backlash-quiet-quitting-fobo-obsolete-white-collar-rebellion/

 

Quote

Goldman Sachs economists reported this week that AI saves workers who use it correctly an average of 40 to 60 minutes per day. The math is almost symmetrical: the productivity AI gives to people who use it well is almost exactly equal to the productivity it destroys for people who can’t get it to work.

That sounds a lot different than we will kick out all employees and let the AI agents do the job.

[rogermunibond]

EQIX and DLR are starting to take off.  quite a few analyst reports on channel checks showing high bookings from corporate clients. this is all deployment of AI/LLM to corporate business processes.

[flesh]

49 minutes ago, frommi said:

https://fortune.com/2026/04/09/ai-backlash-quiet-quitting-fobo-obsolete-white-collar-rebellion/

 

That sounds a lot different than we will kick out all employees and let the AI agents do the job.

It's interesting isn't it? It's still early days though so that supposed balance will naturally skew towards those who can use it. Over time more will learn, those that can't will fall to attrition. I find it fascinating how different prompts, multifaceted prompts, prompts with more qualifiers, and prompts asking for what your missing or not questioning that you should question seriously change the tokens. What your capable of asking is becoming marketable like never before. 

[Milu]

SAAS stocks really being taking to the woodchipper today. 

[Eldad]

2 minutes ago, Milu said:

SAAS stocks really being taking to the woodchipper today. 

Conversely, the Constellation Software group of companies appears to have run out of sellers. 

[Longnose]

20 minutes ago, Milu said:

SAAS stocks really being taking to the woodchipper today. 

Indeed...

 

Many of them are already at crazy valuations...  

 

Ive been doing a ton of playing with AI for coding and really struggling to believe that AI cuts these big boys off at the knees like this. 

 

Im struggling to find a good reason why the whole sector is selling off today too. Its like some whale fund, ETF, or index almost has some forced selling on the sector.  As im not really finding any news or anything.