Believe it or not, there’s a lot of fake AI out there. Smoke and mirrors. Just as we saw with blockchain a few years ago and the “internet stocks” back in the tech bubble… Same song, different verse: some companies will do anything to boost their stock price, including trying to link themselves to a hot theme regardless of whether that link has merit.

Fake Accounting in AI: SMCI…

News flash, not all AI is real, and falling for the wrong AI can crush your portfolio. For example, Super Micro Computer, Inc. (SMCI) was down over 30% this week after its Auditor resigned. Turns out the AI company might have some fake accounting making earnings look better than reality.

Did you know we Suspended the stock and warned of the major risk in “Weak Internal Controls” at Super Micro Computer back in August? We saw an article about accounting troubles and alerted clients on our website as well as those that attended our trainings with Prosperity Pub. In several meetings with clients and webinars over the last few months, we’ve been very clear to point out that our Stock Rating on SMCI warned of the risk in the valuation – despite the very strong earnings the company was presenting to the outside world. We also, of course, pointed out the fact that, since late August, we suspended our Stock Rating on SMCI after the company announced it had to delay filing its 10-K

“to complete its assessment of the design and operating effectiveness of its internal controls over financial reporting” – source Bloomberg.

Again, not a good sign…a big red flag when you see a company say something like that.

Footnotes Reveal Fakes All the Time – But You Have to Look

Note that we usually flag several companies every quarter for auditor warnings, e.g. “Weak Internal Controls”, during filing season, which is when companies file their 10-Qs and 10-Ks with the SEC. We’ll be updating thousands of company models with the latest financial data from these calendar 3Q24 filings over the next few weeks. So, be on the look for timely warnings like SMCI. These days, you never know what you’re going to find in the footnotes! Details on findings from prior filing seasons are here.

But, let’s get back to AI. It is not magic. There is no such thing as magic, especially in the stock market. If you want to believe in magic, I cannot stop you, but I will continue to advocate for doing proper diligence.

Let’s Get Real About AI

Want to know how real AI works? In my interview at the Stansberry Research conference, I get real about AI and what makes it work – like really work, not the smoke and mirrors you see from most AI. I also share how we built our Robo-Analyst technology, the same technology featured by Ernst & Young, Harvard Business School, MIT Sloan School of Management, and the Journal of Financial Economics.

Figure 1: My Interview on Why AI Is Not Magic at the Stansberry Research Conference

Sources: Source: Stansberry Research on YouTube

If you want to see exactly what I am talking about in the interview when I mention the ability to audit our data and know exactly where we get every data point, see this tutorial. It demonstrates our Marked-Up Filings feature, which allows clients to audit every single data point in our models and reports.  As far as I know, no other research firm in the world can show you exactly where or how they get data from the footnotes, but we can.

Not sure if you can trust our numbers? No problem. We give you the ability to audit every single number. If that does not show confidence in our data and models, I do not know what does. Also, the fact that we can offer auditability means that other firms can do the same. If we can do it, why can’t they do it? 

Figure 2: Marked-Up Filings: Unrivaled Auditability of Valuation Models

Sources: New Constructs

Here are the key AI warnings from my Stansberry Conference interview, summarized to help you detect fake AI and quickly assess whether any AI you encounter is real and trustworthy.  

  1. AI is not magic and anyone claiming such should be avoided. Red flag. Turn and run, as you would from SMCI right now.
  2. AI models are only as good as the data that drive them. Bad inputs = bad outputs.
  3. To trust an AI, you must be able to audit the data on which it is based. There is no other way to be sure. If the provider of the AI will not let you see the underlying data driving the AI, see point #1 above. Turn and run.
  4. No reliable AI is based on the internet, e.g. Chat GPT, because there is no way to audit the entire internet.
  5. Reliable AI must be based on a very specific dataset that is properly curated for a machine to be able to properly digest and understand it.

Here’s how we describe our AI:

Our AI-driven Robo-Analyst Process applies machine learning and other technologically-driven methods to a highly-curated and proprietary dataset that we built from carefully marking up 200,000+ 10-K and 10-Q (and similar) filings over the last 20+ years. Highly trained experts marked-up the filings with the goal of categorizing financial data according to a very specific taxonomy. This process allowed us to train machines to replicate the mark-up process, which also allows us to automatically collect financial data, build financial models and produce stock ratings. We provide more details here with graphics that help explain our process.

The reason I can talk about AI with authority is because I dedicated my professional life to building AI over the last 25 years. In fact, I have a United States Patent (#7,752,090-filed in August 2003) for a System and Method to Reverse Accounting Distortions and Calculate a True Value of a Business. This patent shows my early thought leadership in how to use technology to perform fundamental analysis with integrity at scale. The only way to get a patent on something like this is if it’s never been done before. So, I guess you could say I am, at least, among the first to use machines to read filings.

In case you want to know how good New Constructs is at getting machines to read filings, I submit to you:

Does our AI Pay?

Yes, it does. Our novel alpha is not just proven in papers published in top-tier peer-reviewed journals. We have a strong stock-picking track record. Loyal readers know that we’ve been ranked #1 in multiple stock-picking categories, including “All-Time” by SumZero nearly every month over the last three years. SumZero is a community of ~16,000 buy-side only investors; so, we’re going against real competition. “Buy-side only” means professional stock pickers at hedge funds, not Wall Street (i.e. sell-side) analysts.

If you’d like multiple real-world examples of how we’ve found some of the best stocks over the last 15 years, like picking Nvidia (NVDA) at $0.57/share in September of 2015 (see the report), then join my next training session. It’s happening in a few weeks, Monday November 18th at 1pmET. Register here. In case you see this letter after November 18, the replay will be in our online community here. You can join our private community by completing this form.

In that training, I’m not only going to share some of our research and proof that it works, but I will also reveal how we use AI to produce the best fundamental research in the world.

Who Else Uses Our AI?

Beside our thousands of followers and customers, Bloomberg is a big user. In fact, they like our AI so much they decided to create a family of indices based on our AI. The first three are in production:

  1. Bloomberg New Constructs 500 Index – think of this as an enhanced S&P 500 index. Instead of weighting the 500 stocks based on market capitalization, the index weights them based on our proprietary Core earnings. Click on the link to see how it performs compared to the original S&P 500 – you’ll be impressed.
  2. The same is true about the performance of the Bloomberg New Constructs Ratings VA-1 Index, which tracks all the stocks with our Very Attractive Rating in the Bloomberg 1000 Index.
  3. Bloomberg New Constructs Core Earnings Leaders Index is probably the one you’ve been hearing about. It’s already getting lots of attention – see this press release. This index tracks the performance of the top 100 companies that have high Earnings Capture based on our proprietary Core Earnings, which removes unusual gains and losses found in the footnotes and MD&A in company filings.

The bottom line: there is no substitute for hard work and attention to detail when it comes to anything in life, but especially when it comes to AI and machine learning. And, to build AI that produces novel alpha takes an extraordinary team making an extraordinary effort – which is exactly what New Constructs has done over the last several years.

There are no short-cuts for making AI that truly adds value.

Want more details on New Constructs?

We regularly review our work and research on Long Ideas and Danger Zone Ideas with clients. We want you to know how much work we do! Here’s some ways to keep in touch with us:

  1. Free live Podcast every month. The next one is on Friday, November 15th, register here. The last one was October 18th. Get free replays in our online community.
  2. Join our online community (use this form to sign up for free). Ask questions and make friends! Lots of humble investors talking to each other is a good thing.
  3. Monthly Let’s Talk Long Ideas webinars where we do deep dives into our research, analytics, reverse DCF models and ideas for our Professional and Institutional clients. Our next one is on November 12th at 5:00pmET. Replays are here for our Professional and Institutional clients.

If this message resonated with you and you want to start your investing future with us – schedule a meeting with us here.

Diligence matters,
David

This article was originally published on November 7, 2024.

Disclosure: David Trainer, Kyle Guske II, and Hakan Salt, receive no compensation to write about any specific stock, sector, style, or theme.

Questions on this report or others? Join our online community and connect with us directly.

Click here to download a PDF of this report.