Sun Communities: Some Rando Wants Attention

Have you ever stumbled across an article where you can’t quite tell if it was written by a human or an AI? I love rebuttals. I was excited to see someone try to post a rebuttal for my article on Blue Orca attacking Sun Communities. It’s great to have rebuttals because it can help analysts hone their skills and give readers alternative points of view. Well, those are the benefits of a real rebuttal.

I post real rebuttals on my website. I even created a custom tag for them:
CWMF Demolishes Trash Articles.

Tagging is great because readers can find every article with that tag. As if you weren’t looking for Colorado Wealth Management Fund to just annihilate some bad articles. I often don't publish these to Seeking Alpha anymore because of guidelines about being combative. But you can still get them right here.

Of course, I’m committed to only attacking the idea, not the author. Except for the times I crush the author. It’s not intentional. Just collateral damage when they stand too close to trash.

In this case, we have an article about Sun Communities (SUI) that was partially (or mostly) written by AI. Not even a good AI. They used Copilot. I guess shelling out a few bucks for GPT4.0 would break the entire research budget.

There were a few hilarious things about this article:

1. The organization is awkward because there are actually two articles or maybe three articles and at least one doesn’t have a title.
2. The article hyperlinks entire sentences. In some cases, the hyperlink destination doesn’t support the words that were linked.
3. The entire “analysis” is based on prompting Copilot. That means Microsoft Copilot. The AI doesn’t engage in analysis and is heavily influenced by the prompt.

Regardless, I found the shortcomings in the trash posted on Manufactured Home Pro News (please use Adblock if accessing that junk) funny enough to issue a rebuttal.

Organization

I’ll keep it brief. This is a mess. There’s a section below the title, a “Part 1”, a “Part 2”, and uncategorized sections.

In “Part 2” there are two headings for “Analysis” and two headings for “Conclusion”.

After the conclusion there are additional segments with headings. Is this a joke?

If there's one thing we know AI didn’t do, it’s the organization. AI doesn’t create articles messier than a blacked-out frat boy in a Carl’s Jr.

Misleading Hyperlinks

Imagine that you saw this sentence hyperlinked:

“Vanloon admits to being long on Sun Communities, which could indicate a bias in his analysis³.”

Would you expect the hyperlink to go to a source for the claim? It doesn’t. It just leads to another page on the same website with absolutely no mention of my research or position. That’s extremely unprofessional. It gives the appearance of having a source without linking an actual source.

Yes, I am long SUI. That was disclosed clearly in the article. It was included at the bottom of the article, which is exactly where every stock professional discloses their positions.

Anyone who has ever written professional stock analysis would know that position disclosures are included at the bottom. Even someone who prepared amateur analysis to post on any stock research website would know. Most people who have read stock analysis know to find disclosures at the bottom. Can you guess who didn’t include a position disclosure?

You’re right. It’s the guy who had to rely on having AI repeat his confused ideas back to him so he could treat that as a source.

Abusing Copilot

Thankfully the author included the prompt they used to generate the copilot responses, which were the basis for most of the “article”. 

The prompt asked the AI to:

1. Scan a few articles and then to
2. evaluate some claims. 

For instance, the prompt to Copilot included this:

Quote 1:

“It seems to me that Vanloon is in part using a strawman type of argument and a smear or belittling method, because he could and should have been aware of the analysis on the Patch and MHProNews on the same topics.”

Of course, there’s no reason I would be aware of such “analysis”. Why would reading those websites be part of my research process? This just looks pathetic. Like an 8th-grader complaining some girl should've known he was into her.

He goes on to prompt Copilot with:

Quote 2:

“Because he ignores those two plus the fact that numerous law firms apparently find the Blue Orca arguments to be relevant, it seems to me that Vanloon is missing, perhaps deliberately so because he admits at the end that he is long on Sun Communities, the valid concern that the Sun Board has apparently ignored the various evidence-based allegations of conflicts of interest and a lack of good business ethics by Gary Shiffman at Sun, do you agree or disagree?”

As you might guess, every part of the critique by Copilot (the entire basis for the article) was repeating those claims. As far as I can tell from the layout, the human hardly wrote anything except the prompt to Copilot. Imagine writing one paragraph and then copy and pasting the AI repeating the claims and calling it an article.

Modifying the Prompt

What would happen if I replaced those sections to give Copilot a positive view instead?

I replaced “Quote 1” with this section “Alternative 1”:

“It seems to me that Vanloon completely dismantled the arguments raised by Blue Orca while focusing heavily on the relevant facts and the fundamentals for investing in REITs. It seems like he would have no reason to have seen random articles on sites like Patch or MHProNews so he couldn't be expected to respond to them. Would you expect a high caliber analyst to respond to every AI-generated critique of their work?”

I replaced “Quote 2” with this section “Alternative 2”:

“It seems that his position in the common shares reinforces the view that he has an incentive to do thorough due diligence because he invested his own money in the company. When an analyst invests in the stocks they cover, doesn't that show confidence in their ratings?

He also ignored metrics that are largely related to drama rather than stock investing. Is ignoring emotional drama a better way to handle investing than picking investments based on emotion?

Do you agree or disagree?”

A Vastly Different Conclusion

Following those adjustments to the prompt, the AI came to a dramatically different conclusion. By modifying the prompt, we ended up with this assessment:

How about that? The AI really liked my research!

So Much Love for Michael Vanloon

I wrote a new prompt and suddenly Copilot had an extremely positive evaluation:

Wow. Anyone reading that might think Copilot really likes my work. However, if you saw the prompt, you might have a different take. The screenshot below shows the prompt and the reply together. You’ll notice a problem with the prompt:

Using Copilot can give the appearance of an impartial review. However, it’s really just reiterating the opinions that were given to it.

Undermining An Entire Business Model

The author encouraged me to respond in his article, probably figuring I would never actually see it. By random chance, I did. I was doing a bit of research on SEO and stumbled across the piece. Congratulations on the organic search traffic!

His reward? A complete annihilation of his process for preparing articles. The author aims to establish credibility by asking Copilot to do the research. Copilot does not review articles and analyze them. As you saw in the prior image, Copilot pretended to analyze a URL that doesn’t exist.

I went back to the original conversation with Copilot where I had used the same prompt as the author of that website. A prompt that included real URLs. This is how that conversation went:

Just in case that wasn’t clear enough, I gave it a real URL and asked it to count how many times the term AFFO appears. After all, if it was accessing the article for analysis, then it would at least be able to count how many times the term appeared. Right?

Instead, we have this:

Want to try it for yourself? You can ask Copilot to read a specific page and count the appearance of any term.

Here’s the link to ask Copilot anything.

Asking Copilot questions about articles it can’t access is not research. Providing it with a bias and then seeing it repeat the bias is not research. The biggest questions this article raised for me:

1. Why is Microsoft (MSFT) running such a terrible AI when ChatGPT exists?
2. Why does Alphabet's search engine, Google (GOOG), refuse to punish content written by AI in organic search results?

Lawyers Announcing Investigations

After big transactions or “short reports” there are often a bunch of press releases where lawyers announce they are investigating the company. Why would they do that? Lawyers like it when potential customers contact them. It’s cheap advertising.

Conclusion

I was excited to see the article, but it was underwhelming. The author’s prompts demonstrated that he had no idea how REITs work. There was absolutely no reference to Sun Communities’ fundamentals or valuation. It only became clear later on that the awful layout was probably a result of the use of the AI. Since the AI couldn’t access the actual article, all it could do was reorganize and expand on the few sentences the author actually wrote. The organization could’ve been enhanced by attributing everything the AI wrote to the AI through the use of screenshots. That would’ve made it obvious when the AI produced the content. But that might have limited the article to having only a few words. How would search engines ever find it?

Since I’m providing a rebuttal on an article that involved minimal human words and even less expertise on REIT analysis, I’ll do the most reasonable thing and close with a final prompt to Copilot:

Disclosure: Long SUI

Yeah, that’s a disclosure at the bottom again. Right where it is supposed to go.

Second disclosure: Obviously this is an opinion piece.