Are LLMs enough? They’ll have to be, for now

What’d you say if LLMs are as good as AI gets, at least for now? If AI isn’t the magic key to all questions but more like, say, a Swiss Army knife?

The coolest Swiss Army knife, the one with the scissors and the magnifying glass and the ballpoint pen. But without the brain.

Then what’d you say? I’d say fine. A lot of people would say fine, many of them right here at LinkedIn. Hell, even a lot of the top minds in AI would say, well, maybe not fine, some other F-word, but they’d live with it.

They’re living with it now, because the quest for that magic key isn’t going to plan.

The plan, inspired by a 2019 essay titled “The Bitter Lesson,” by computer scientist Richard Sutton, was to reach god-mode AI by throwing galactic amounts of compute at general-purpose models, along with trillions in capex, entire regional water tables and enough power to run a midsize nation for 10 years.

A lot of smart people bought in: Altman, Amodei, Zuckerberg, Musk.

But other smart people opted out: LeCun, Sutskever, Murati--people who think the way to AGI is not by brute force but by theoretical breakthroughs.

And it’s turning out they’re right. Not because we’re on the cusp of those breakthroughs but because the brute-force method is running into even more brutal technical, financial and political realities.

Which brings us back to the Swiss Army knife.

If autonomous AI isn’t going to replace human intellect in the foreseeable future, then the best we can do for now is LLMs.

Which, honestly? Is completely OK, especially if you’re a leader looking to lift your profile and demonstrate your expertise.

For that, an LLM is a great tool. Let it handle the research and the rough draft, then pass the draft to a human to add nuance, originality, personal expertise.

That’s the stuff the Google and LinkedIn algorithms want now and that’s the stuff your Swiss Army LLM, cool as it is, can’t do.