Post New

I'm still trying to wrap my head around the nuts and bolts of the AI build out.

Yesterday, I learned that China is developing its AI "in the cloud" -- some of which is provided by Amazon and Google and some of which is provided by Chinese electronics companies for national security reasons. This allows the AI developers to avoid spending capital on their own data centers.

Different companies provide different aspects of the AI build-out. The AI bubble will affect them differently. While Amazon and Google have businesses that produce positive cash flow the companies that have to borrow to build data centers are suffering from investors' recognition of the bubble.

https://www.wsj.com/tech/ai/coreweave-stock-market...

CoreWeave’s Staggering Fall From Market Grace Highlights AI Bubble Fears
The data-center provider’s terrible six-week slide picked up speed when a famous short seller piled concerns on top of delays

By Robbie Whelan, The Wall Street Journal, Dec. 15, 2025

....

CoreWeave’s business model involves using high-interest debt to buy thousands of advanced AI chips from Nvidia, installing them in server racks inside data centers that it leases from third-party landlords, then renting access to the chips to AI companies. ...

Some critics point to the high levels of debt it has taken on to finance its data-center build-out, while others worry that the company depends on just a handful of large customers, such as OpenAI, Microsoft and Meta, for the bulk of its revenue. CoreWeave saw sales more than double in the most recent quarter to nearly $1.4 billion, from $583 million a year earlier, but the company is unprofitable and lost $110 million in its most recent quarter...

The rapid pace of growth has raised questions about how and when major capital investments are going to produce healthy profits. They also underscore the intensity of the frantic rush to build enough computing infrastructure to satisfy demand.... [end quote]

https://www.cnbc.com/2025/12/15/ai-infrastructure-...


AI infrastructure selloff continues on Wall Street as Broadcom, Oracle shares slide
by Ari Levy, CNBC, Published Mon, Dec 15 20252:58 PM ESTUpdated Mon, Dec 15 2025

Shares of Broadcom, CoreWeave and Oracle all dropped on Monday, extending last week’s declines.
While the artificial intelligence boom shows no signs of easing, investors have grown concerned about how it will all be financed.
Oracle has lost almost half its value since the stock peaked in September.... [end quote]

The "Magnificent 7" represent 40% of the value of the S&P500. How will this affect them?

Wendy

Yesterday, I learned that China is developing its AI "in the cloud" -- some of which is provided by Amazon and Google and some of which is provided by Chinese electronics companies for national security reasons. This allows the AI developers to avoid spending capital on their own data centers.

Wendy,

It is interesting when put on a timeline against TikTok's marketing. TT was snooping on all Americans to find out who we were. Supposedly. But now with AI making leaps and bounds, does all of that prior fluff does it even matter?

The central feature of AI is search. Search today using AI is so much better than getting some half-informed blogger. Citations are needed by the new system to present the information.

The other central feature is FSD. Clearly, the accident rate will descend dramatically.

The central feature of AI is search. Search today using AI is so much better than getting some half-informed blogger.

I certainly agree that it is “easier”, but the reality is that it is also less accurate than going to an authoritative source.

If you know something about the topic you’re searching, you won’t be surprised sometimes to find mistakes, poorly drawn statistics, or outright hallucinations included in the summary. And rather than go to a good source (and yes, there are half-informed bloggers too) most people will just skim the result, anccept it (because Google has been pretty good in the past) and move on.

“Easier” is good, but “Accurate” is a whole lot better.

Maybe they’ll figure out a way to fact-check the fact-check, but at the moment they don’t, and everyone is liable to accept whatever the machine spits out. That’s not really “better” by any measure.

If you know something about the topic you’re searching, you won’t be surprised sometimes to find mistakes, poorly drawn statistics, or outright hallucinations included in the summary.

I'll second that. AI search is basically useless for most things I have to look for professionally. When I look for zoning information, it almost always gets the city wrong. Whatever I'm looking for, it will return the City of Miami requirements or code provisions - even though I'm usually looking for a different city. I guess it's just because more people live in the City of Miami than cities like Miami Shores, West Miami, North Miami, North Miami Beach, or Miami Beach. So AI gets the question wrong - and if you rely on it if it tells you that your setback is X or your lot coverage is Y%, you'll be badly misled.

There is a way to fact-check the response. Plug the response as a prompt into a different AI and tell it to validate the information from authoritative sources in the specific location only, not social media or make assumptions from inferences.

While there are many many things that might pop the bubble, this was interesting:

https://archive.ph/oxMSl

We’ll begin to see weakness in the AI story after the fact, whether from cancelled orders or failure to convert technology into profits. Will short-sellers be able to spot the reversal in time to place some quick bets and turn a profit? Yes, says Mr Left, “but only after missing the first 15-20%.” Then, the main sign that a prolonged bear market has begun would be companies cutting spending, or “big firms saying they’ve experimented with AI and it’s not been transformational.”

For Mr Block, meanwhile, the key indicator that the cycle has turned would be a sharp rise in unemployment. He thinks a crucial driver of the bull market has been the monthly flow of retirement savings into passive index funds, which buy shares in the biggest firms regardless of their price. An abrupt rise in joblessness would reduce that flow, and with it support for share prices to keep rallying.

Whenever the next bear market begins, there is no shortage of reasons to worry about its impact. Mr. Left points to the radical uncertainty over how AI will develop, and which companies will end up exploiting it most profitably, as a significant source of potential losses for shareholders. Then there is the question of who, eventually, will end up paying for firms’ vast AI-related capital expenditures. Big tech firms, for instance, are set to invest $5trn between now and 2030; JPMorgan Chase, a bank, estimates they will need additional revenues of some $600bn a year to make these worthwhile. “It could [come from] businesses or consumers,” says Mr. Left, “so which one is it? Do you really want to be the person trying to guess?”

The second reason to fear a bear market is that American households have more to lose from one than ever before. Stock ownership now accounts for over 30% of their total assets, the highest since the Federal Reserve started collecting such data in 1945 (see chart 2). At the peak of the dotcom bubble the equivalent figure was 27%.
Moreover, Mr. Chanos sees evidence that retail traders have concentrated their bets on the riskiest assets. He notes that the average account on Robinhood, a popular trading app, holds around $12,000. After adjusting for deposits and withdrawals, he estimates that account values dropped by 7% in October and November—a period during which the stock market wobbled, but ultimately recovered. That balances nevertheless fell suggests they were “way out there on the risk spectrum”, invested in stocks that dropped more and recovered less than the wider market.

Just remember - it's only a game :-)

Jeff

When I look for zoning information, it almost always gets the city wrong. Whatever I'm looking for, it will return the City of Miami...

Frankly, I'm surprised at that: that when you are specifically asking for Miami North Shore or West Miami (or whatever) you get City of Miami responses. My own experience is that, while not perfect, if you are asking specific questions with detail and go back and forth with prompts to elicit better answers with more specifics, you get a much better and accurate answer. I agree that you have to check **everything** (why wouldn't you and I would do this even with The Economist or Encyclopedia Britannica - if that still exists), but I've been getting great answers for a couple years now. I also agree that plugging the same questions and also responses into another AI search tool as flyingcircus says, often works well, too.

Pete

Frankly, I'm surprised at that: that when you are specifically asking for Miami North Shore or West Miami (or whatever) you get City of Miami responses. My own experience is that, while not perfect, if you are asking specific questions with detail and go back and forth with prompts to elicit better answers with more specifics, you get a much better and accurate answer.

Nope - I'll get the Miami-Dade or Miami substantive responses, often misattributed to whatever City. The live links will also go to the big jurisdiction code provision. The AI box from google is the worst. I just googled "South Miami setbacks," and got this response:

South Miami setback rules dictate minimum distances for buildings from property lines (front, rear, sides) and vary by zoning (like RU-1, RU-2) and lot specifics (corner lots), generally requiring significant space (e.g., 15-25 ft front, 10% lot width for sides) but allowing some flexibility for sheds or pools; you need to check your specific zoning (Miami-Dade Code) for precise requirements, as they ensure light, air, and fire safety.

Very little of that is right. South Miami doesn't have zoning districts labeled RU-1 or RU-2 (those are county zoning designations), their front setbacks aren't 15-25 feet and they don't use a percentage of lot width for side setbacks, you shouldn't check the Miami-Dade Code for precise requirements (municipal zoning trumps the County requirements), and all of the links that are provided are to either the County or the City of Miami zoning regulations (the City of Miami is a separate city from South Miami).

I suspect it's an instance of what I think is a general principle of AI - namely, how it handles a subject is only as accurate as the degree people use the internet to talk about the thing. People don't write about the arcana of zoning rules the way that they do about Harry Potter. There isn't a massive repository of writing samples about zoning like there is Github for coding. Etc.

Because there's less material for the model to learn from, the mistakes are more egregious.

South Miami setbacks

Here’s a **clear overview of setback requirements in the *City of South Miami, Florida*** — i.e., the minimum distances structures must be from property lines — based on the latest **South Miami Land Development Code** and related zoning provisions: ([Municode Library][1])

---

## 🧱 What *Setbacks* Are

A *setback* is the **minimum horizontal distance** a building or structure must be from a **property line or right-of-way**. They’re used for safety, privacy, aesthetics, and infrastructure planning. ([Wikipedia][2])

---

## 📏 South Miami Setbacks — Residential

In **single-family residential districts** (e.g., RS-1 through RS-5), setbacks are defined in **Section 20-3.5** of the Land Development Code and vary by zoning category and lot type: ([Municode Library][1])

### 🏠 **Minimum Yard Setbacks**

* **Front Yard:**
Minimum distance a house must be set back from the front property line varies by zone. For typical single-family lots it’s commonly **25–35 feet** depending on the district and lot size. ([Municode Library][1])

* **Rear Yard:**
**At least 25 feet** from the rear property line on many residential lots. ([Municode Library][1])

* **Interior Side Yard:**
Minimum side yard setback varies but is generally greater for larger lots (e.g., ~7.5–12.5+ feet) and is related to lot width and zoning category. ([Municode Library][1])

* **Side Street (Corner Lot):**
Side setback on a street-facing side of a corner lot is typically **15–20 feet** depending on the district. ([Municode Library][1])

📌 *Note:* There are **different tables** within Section 20-3.5 for **one-story vs. two-story homes**, and second-story side street setbacks may be larger to maintain neighborhood character. ([Municode Library][1])

---

## 🛣️ Supplemental Setback Rules

### 🚗 Roadway and Right-of-Way Setbacks

* **No structure (except utility poles)** may be located closer to the **road centerline** than **half the official right-of-way width plus the required yard setback**.
Setback distances are measured from the **official right-of-way line**, even if not yet dedicated. ([Cloudfront][3])

* Public right-of-way widths are specified for various roads (e.g., 100′ for Bird Road, 80′ for Kendall Drive, 50′ for other roads) and used to determine setback measurement bases. ([Cloudfront][3])

---

## 🏘️ Measurement and Corner Lots

* **Setback measurement** is from the **official right-of-way line** or property line, not necessarily the curb or paving edge. ([Cloudfront][3])

* At street corners, there are special **visibility triangle rules** where nothing may obstruct view within certain corner triangles (e.g., no plants/structures between 3–6 feet high in designated areas). ([Cloudfront][3])

---

## 📌 Practical Tips

* **Check your zoning designation** (RS-1, RS-2, etc.) to find precise numbers — each category has its own table in the code. ([Municode Library][1])
* **Permits and Variances:** If you can’t meet setbacks for a project, you may need a **variance** from the Planning & Zoning Board.
* **Accessory structures** (sheds, pools, fences) also have their own setback rules that can differ from the main building. ([Miami-Dade County][4])

---

## 📍 Where to Find the Official Rules

* The **South Miami Land Development Code**, particularly **Section 20-3.5 (Dimensional Requirements)**, contains all setback tables and measurement details. ([Municode Library][1])
* For questions about your specific property, the **Planning & Zoning Division** can provide zoning maps and personalized interpretation. ([southmiamifl.gov][5])

---

If you want, I can help you **find the exact setback numbers for a specific address or zoning district** (e.g., RS-3) — just let me know!

[1]: https://library.municode.com/fl/south_miami/codes/... "Land Development Code | South Miami, FL - Municode Library"
[2]: https://en.wikipedia.org/wiki/Setback_%28land_use%... "Setback (land use)"
[3]: https://d3n9y02raazwpg.cloudfront.net/southmiami/a... "CITY OF SOUTH MIAMI"
[4]: https://www.miamidade.gov/global/economy/zoning/di... "RU-1 – Single-Family Residential Zoning District"
[5]: https://www.southmiamifl.gov/170/Planning-Zoning-D... "Planning & Zoning Division"

In **single-family residential districts** (e.g., RS-1 through RS-5), setbacks are defined in **Section 20-3.5** of the Land Development Code and vary by zoning category and lot type: ([Municode Library][1])

and vary by zoning (like RU-1, RU-2) and

Al,

"South Miami doesn't have zoning districts labeled RU-1 or RU-2"

AI is not saying they do have RU-1 or RU-2. Like or eg.

The problem with most AIs is that they are convinced about the correctness of any statement they make - even if the source of that data is dodgy. If you point out the error of their ways, they will compliment you on your astuteness, but not apologize for sending you down a rabbit hole. The best way to check their work is to have another AI try to tear it apart.

I'm anticipating the day when retailers (et al) will use their own AIs to influence AI shopbots - much in the way that resellers currently try to game Google's algorithms for example.

Jeff

AEO

AI engine optimization

I just began to use it. It will take a year to matter.

The idea of bot AEO is so so. AI works against AI materials because of hallucinations.

None AI materials go to the front of the line.

Because there's less material for the model to learn from, the mistakes are more egregious.

There are some questions that have no definitive answers. Yet, the AIs try to answer anyway resulting in gibberish. Heck, "zoning" has no definitive answers because "zoning" is almost always negotiable. If there is a certain setback in an area, and a big developer comes along, ready to invest million, tens, hundreds, of millions, etc and requests a variance, they will probably get that variance (unless it really egregiously affects something real).

If you know something about the topic you’re searching, you won’t be surprised sometimes to find mistakes, poorly drawn statistics, or outright hallucinations included in the summary.

I'll second that. AI search is basically useless for most things I have to look for professionally.

From just looking at basic tax law, it's spot on, does a good job. I was given this basic book to help me plug into the software and, found that if I queried refundable credits, the list was much better than the one in the book, though not complete, it was more complete. And I can query how to input into my software, and it did a good job.

I find I can get different answers depending on how I ask the tax question, but they're all pretty good responses. So what are all the different AIs? I'm only familiar with Google.

If you know something about the topic you’re searching, you won’t be surprised sometimes to find mistakes, poorly drawn statistics, or outright hallucinations included in the summary.
...
I'll second that. AI search is basically useless for most things I have to look for professionally.
...
From just looking at basic tax law, it's spot on, does a good job...

One thing I find them good at is summarizing an existing document or corpus, perhaps explaining the good tax checklists. There is no need to make stuff up or attempt reasoning, so they don't.

Beyond that, check *everything*. I like the way Perplexity gives footnotes for what source it's citing for each statement, but more and more they seem to be web pages which are themselves either AI slop or "some guy on Reddit".

The typical LLM is like a person who knows a truly enormous amount of stuff having read absolutely everything on the web, but unfortunately believed it all, and has an imperfect memory to boot. Ask for the most reliable hammer and it will either cite a brochure from a hammer manufacturer, or recommend a very good tool company that doesn't offer hammers.

Jim

Beyond that, check *everything*.

Yes, for sure.

Also, what a lot of folks do not do is be very specific in their " ask" and to not ask follow-up questions or add more information to an AI conversation. It really is all about the prompts and follow-ups! I have had very good experiences over with AI the past few years by following that rule.

One youtuber posted this, which may be helpful to some as example:
https://www.youtube.com/watch?v=Ks__4nZbOWE ,

Pete

Post New

US Policy FAQ

---

Contact Shrewd'm
Contact the developer of these message boards.