You could mess around trying to exclude big tech like some wacky S&P 500 ex-tech fund or try to overweight small and mid cap. That’s if you absolutely must hold US though.

For me about 70% of my portfolio is a world index with overweight on international. 60% VTI + 40% VXUS or equivalents. This is considered overexposure to international by conventional US investors.

The other 30% is all-in on China via MCHI. I’m edging towards increasing international and China exposure and reducing US. The US large caps are fundamentally overvalued so I’ll probably do this soon. Just hard to make that mental jump even though the material analysis says so.

To be clear, this is a risky strategy. All the major players and investors worldwide are adamant there is no AI bubble, and no US bubble, and China will never pay off or grow.

It’s just about what you’re willing to risk to say you’re right and they’re wrong. Safest choice is VT and chill.

I work in tech in one of those companies that thinks AI can solve literally everything, and what I hear from leadership on these kinds of reports is that we just haven’t figured out how to measure the success yet. So probably it won’t be the trigger.

The writing is on the wall though, a couple of weeks ago an exec came flying in with this great idea for how we can replace an entire product with AI. Honestly we probably could, it wasn’t a bad idea. The issue came in with the cost…even at the heavily subsidized by venture capital cost, it was going to cost us more per customer than what they pay us, just to do this one small part of the overall workflow, using AI. So sure, we could build out to this AI that does all this cool shit for us (except when it goes off the rails and does it wrong, which definitely happens). But the cost was prohibitive NOW. Imagine how expensive it will be when the real costs start propagating down to the buyers.

What I’ve learned about AI working adjacent to it is that it actually can do quite a lot, as long as you don’t need it to be perfect and as long as the real total cost of using it is actually less than writing real code that does the same thing. Which is not gonna apply to the vast majority of dumb things people are trying to do with AI. That’ll be the real trigger; shit goes off the rails because the AI can’t do it predictably and/or as the true cost starts becoming apparent nobody can afford it anymore. And all the companies who are reliant on it will collapse entirely.

Its not a coincidence that suddenly a lot of banks have been issuing warnings over the AI bubble. They already know.

looks like this is from january, so I guess it would have if it could have. Sites like HN are full of “but have you tried it THIS week??? Otherwise your opinion is invalid”

I don’t know what to expect from a correction since there’s just constantly a “new, better model” every day, thousands on hugging face alone, let alone proprietary ones. It makes the effort to prove the negative (ie. These models are not useful for most tasks) very high so I don’t typically argue with people about it. I think that the constant treadmill to keep up with the “improvements” is part of the marketing, to be honest.

I’m sure this is typical of every bubble but to be honest I’m not sure what’s going to pop it if GPT-5 didn’t, if OpenAI’s P/L statements don’t, if businesses aren’t fatigued yet with all of the “news” about new models, if the economics not making sense hasn’t stopped anyone… it’s a weird time in tech right now.

LLMs are a useful tool for a bunch of things, but it turns out they’re not really easy to monetize directly. I’ve also noticed that the approach in China is completely different where LLMs are just treated as infrastructure you build stuff on top of. What we see in the west is a bubble akin to what we saw during dotCom days, where investors are throwing money at this stuff and people are building all kinds of dumb shit that doesn’t actually make any sense.