Not so fast…

As hard as it may be to fathom, there are people who know stuff, and still read this pathetic blog. Let’s hear from a couple of them.

Days ago Dr. Garth responded to a real doctor wondering if he should dump his medical professional corp and join the defined-pension pension plan run by HOOPP (Ontario’s well-respected asset manager). We reviewed a few of the ways government is hoovering these corporations, and said going the DP route was a viable option to consider. After all, the feds want well-paid medical people to be taxed just the same as everybody else (like poor financial advisors), whether they’re incorporated or not.

Well, Jean-Pierre Laporte wrote me with what he thinks is a better idea. He’s the CEO of Toronto-based Integris Pension Management.

“Was looking at your article that mentions that physicians who are incorporated in Ontario can now join HOOPP,” he says.

Doctors should not join HOOPP if they can set up a Personal Pension Plan for themselves.  While HOOPP is well managed, it suffers from the ‘actuarial gain’ problem on death:  your capital stays in the central pension fund to support the retirement of other doctors who are still alive, because HOOPP only guarantees payments during what is called the “guarantee period” which is normally 5 years from the date of retirement (15 if you are a single person and can’t give your surviving spouse a survivor’s pension).

Under a Personal Pension Plan, the same surplus that arises because of an early death is controlled by your Medicine Professional Corporation who can pay it to the surviving children, or return it to the company, or if a family member still accrues service in the pension plan, continues to grow tax-sheltered for very long periods of time.

JP has a valid point. We did not comment on PPPs, which was an omission. For doctors (and other self-employed, incorporated professionals that other salaried Canadians love to hate) this is an option with legs.

Now, let’s talk about AI. The Mag 7, billions (soon, trillions) in data centre spending, the Nvidia miracle and investor giddiness have propelled stock markets to multiple historic highs this year despite Trump, tariffs and Bad Bunny.

But is it all a trap?

“I’ve been enjoying your blog posts as usual, but I wanted to mention: you might want to consider being somewhat more skeptical about “AI”, because many of us in the tech industry who understand a bit about how it works — but aren’t invested in it working/aren’t part of an AI startup — are *very* skeptical whether it can *ever* possibly do the stuff it’s being hyped for (ie: replacing most human jobs),” writes Raigan Burns who – among other things – is the Founder & VP at Metanet Software.

Specifically: LLMs fundamentally cannot tell which statements are true or false. It’s no different than autocomplete: it works in the realm of pure statistical correlation. It can judge how *plausible* a string of words is — how often it has seen those words in that order in its training data — but unfortunately the truth isn’t statistical.

This is a fundamental limitation of the technology and it means that chatbots *cannot* ever be relied on to provide correct information — they’re designed to provide plausible-sounding text. And they’re very good at that — but not many useful jobs require writing that doesn’t need to be correct.

If you ask it about something you don’t know very well, it sounds reasonable — but try asking it about something that you know *very* well, and you’ll see errors. Not to mention, it’s prone to “hallucination” (inventing things). This isn’t something that can be fixed, it’s fundamental to the basic way the technology works (statistical, non-analytical).

This technology isn’t in any sense true AI (now called “AGI” because the hypesters have co-opted “AI”), it’s just the next evolution of machine learning/”big data” that’s been going on since the 90s — and it has a lot of the same limitations. (I took a bit of “computational intelligence” (as it was called back then) at U of T.)

It fundamentally cannot, and will not ever be able to, judge the truth of a statement — this means it can’t correct itself, detect when it’s wrong, or even judge how likely it is to be wrong. This is a big problem when you consider how it’s being touted as a valid replacement for human workers!

There are a lot of hypothetical cool uses for this tech, and image generation is maybe one of them (images that look plausible but are fake are much more useful than information that sounds plausible but is incorrect) — but much like self-driving cars, the actual reality of how it will pan out is vastly different (and much less valuable) than what the marketing spin wants you to believe. It turns out that people typically want their information to be correct rather than merely plausible-sounding!

Anyway — I do think there is *something* useful that this technology can do, but sadly this is not likely to be anything as useful as how it’s being hyped. And also it’s incredibly expensive and resource-intensive. Hopefully the massive crash when investors finally figure this out isn’t going to be *too* devastating.

Gulp. Let’s hope not.

The Magnificent Seven – massively engaged in the development and implementation of AI – now account for a third of the entire S&P 500 capitalization. That index so far this year is ahead 16.5%. Since the Trump-Liberation-Day disaster in early April, the market has climbed an astonishing 37%. During that time AI investment has climbed inti the hundreds of billions. The chief AI chip-maker, Nvidia, became the first-ever $5 trillion company. Even Apple passed $4 trillion.

Investors have been snorfling up tech as never before. Well, not since the 2000 dot-com orgy. Back then, in less than two years, the S&P lost 49%. The Nasdq shed 77%. But that was then. This is now. We all learned a lesson.

Right?

About the picture: “I left a federal crown at 53. Reduced pension, however thankful for the pension,” writes Stuart. “Left 247 sick days on the table. Loved my job in the credit field. Recent experience with Service Canada: takes 6 months to change my mother’s address. Even got a follow up letter to confirm it. Something’s wrong. Here’s a picture of a crowded couch at the Hunt camp.”

To be in touch or send a picture of your beast, email to ‘garth@garth.ca’.