AI Stock Forecasts: Machine vs Human Picks

AI stock forecasts are the contested ground of 2026 investing: quant shops say the machines already win, fundamental managers say the machines hallucinate edge, and the newsletter industry has started selling both arguments at once. Whitney Tilson’s N.E.W. System presentation is a useful specimen precisely because it straddles the line: a machine score with a human veto. We read the presentation end to end. Our full AI Civil War teardown covers it in detail.

The Machine Case, Per the Pitch

The N.E.W. System’s engine is the Stansberry Score, a 0-to-100 rating across 5,300-plus stocks: “we figured out a way of understanding those trillions of combinations and crunching them down into a single number from 0 to 100.” On top of the score sits portfolio simulation: “we then trained our A.I. to simulate the future outcomes of thousands of different types of portfolios using every combination of companies imaginable.” The output is a quarterly 20-stock model portfolio, a position-sizing calculator, and monthly updates.

The claimed results are the backtest: “You’d have outperformed gold, bonds, real estate, and even Berkshire Hathaway. And you’d have turned every $100,000 of your money into $339,000 with less risk than the stock market.” The score-and-outcome demonstrations run both directions, high scorers like Cadence Design Systems (86), Agnico Eagle Mines (89, “up as much as 118 percent”), Nvidia (74 back in 2016), and low scorers like “Resolve AI” at 37 (“would have wiped out 77 percent of your money since the summer of 2024”).

The Human Case, Per the Presenter

Tilson’s own credibility block is entirely human, and it is one of the most verifiable in the business: the CNBC “Prophet” nickname, the $200 million fund (Kase Capital), the 2000 dot-com call, the Bear Stearns and Lehman bankruptcy warnings, the 60 Minutes market-bottom appearance (December 2008), the Lumber Liquidators investigation (also on 60 Minutes, 2015), the 2022 selloff warning, and the New York mayoral run. The pitch’s pitch-perfect line on why share the system at all is personal, not algorithmic: “my mom would regularly tell me that I was among the luckiest people to ever walk the face of the earth, and that if I used my good fortune for nothing more than self-enrichment, she would consider me and herself a failure.”

And the human check is built into the product: “you won’t be blindly following a computer. I still vet each recommendation for you along with a special committee of experts.”

How the Two Forecasts Fail Differently

The honest comparison benefits from naming how each side’s forecast breaks when it breaks. The machine’s failure mode is regime change: a model trained on the historical relationship between fundamentals, positioning, and returns will produce confident, specific outputs right up until the relationship it learned stops holding, and a screener cannot tell you whether you are living through one of those stops. This is not hypothetical; quant strategies built on historical patterns have had well-documented drawdowns exactly when market structure shifted. The human’s failure mode is narrative drift: a framework flexible enough to absorb new information is also flexible enough to rationalize stale positions, and conviction without a model’s discipline can hold a losing thesis long after the data turned. The presenter’s own answer to the second risk is his willingness to put the N.E.W. System’s ratings on public display with named sells like Rivian, which creates a record the narrative cannot quietly edit. That is the genuinely useful design here: not that either forecast is right, but that each one’s mistakes will be visible and attributable, which is more than most forecasts of either kind offer.

The Honest Comparison

The honest read of machine versus human, applied to this product:

What the machine demonstrably does well: breadth (5,300 stocks, scored daily), consistency (no mood, no narrative bias), and symmetric warnings (the sell side gets the same treatment as the buy side, which is where the free picks show their work: Garmin (GRMN) scores high for a data-monetization story most humans file under “watch company”; Rivian (RIVN) scores low for cash burn and dilution most humans file under “EV hope”).

What the human demonstrably adds: the 2000, 2008, and Lumber Liquidators calls were judgment calls, made against consensus, before any score existed. Whether that judgment transfers to vetting a quant system’s output is the bet the product makes.

What neither can promise: the backtest is a simulation (“extensive multi-year backtesting”), the “less risk than the stock market” claim is undefined in measure and period, and the forward claims are hedged with “could” throughout. The honest holding: the machine’s past is code and the human’s past is documented, and the product’s future is neither.

What You Get

The N.E.W. System: quarterly portfolios, monthly updates, the Stansberry Score tool, Stansberry GPT, “Ripping Point Riches” and “Ripping Point Casualties,” and the mystery bonus. Price behind the order button, 30-day full refund with keep-the-reports. Details in the presentation. For the machine, our N.E.W. System explainer covers the mechanics, and for the publisher, our Stansberry Research review covers the house.

The Honest Read

AI stock forecasts versus human picks is the wrong frame for this product, because it sells both and discloses the weld: a machine score with a human veto, which is how most serious quant shops actually run. The honest question is not machine versus human but whether this particular hybrid’s backtest survives contact with live markets, and that answer is only available forward. The frame is interesting. The fine print is decisive.

NewsletterVetter is an independent publication. We receive compensation from some of the services we review through affiliate links. Nothing on this site is investment advice. Always do your own research.