Look, I wasn’t planning on writing tonight. The bottle of Jim Beam was keeping me warm company while I watched reruns of Star Trek, but then this gem landed in my inbox. Ilya Sutskever, the guy who recently tried to push Sam Altman off the OpenAI throne (and failed spectacularly), is now preaching about AI unpredictability. The irony is thicker than the morning-after taste in my mouth.
Here’s the real kicker - Sutskever just figured out what any halfway decent drunk could tell you: there’s only so much bourbon in the bottle. Or in his case, “we have but one internet.” Revolutionary stuff, right? These geniuses have been feeding their AI models with every scrap of data they could find, and now they’re hitting the wall because - surprise, surprise - we’re running out of fresh data to feed the beast.
Between sips of whiskey (purely for journalistic integrity), I can’t help but laugh at the beautiful absurdity of it all. These folks have spent years telling us how their algorithms would consume all human knowledge and spit out digital wisdom. Now they’re standing around like teenagers who just realized the keg is empty at their first house party.
But wait, it gets better. Their solution? Make AI that “reasons” like humans. Because that’s exactly what we need - machines that are as unpredictable as my ex-girlfriend after her third martini. Sutskever’s basically saying, “Hey, remember how we promised AI would be this perfectly controlled, predictable thing? Well, scratch that. The smarter we make it, the more it’s going to surprise us.”
He brings up that famous AlphaGo Move 37 against Lee Sedol. For those who don’t remember (I barely do, and I wrote about it), this was when the AI made a move so weird that experts thought it was a mistake - until it wasn’t. It’s like that time I accidentally mixed bourbon with tequila and somehow wrote my best article ever. Sometimes brilliance comes from places you least expect it.
The whole thing reminds me of last week’s tech conference where I asked the keynote speaker if AI consciousness was just digital delirium tremens. Nobody laughed, but I stand by that question. Sutskever’s talking about “self-aware” AI like it’s inevitable, while most of us are still trying to figure out why our smart thermostats think 80 degrees is room temperature.
And here’s what really gets me: this guy just helped orchestrate one of the biggest corporate coups in recent memory, failed miserably, admitted he regretted it, and now he’s back telling us about how unpredictable the future will be. You couldn’t make this stuff up if you tried. Though God knows I have, usually around closing time.
The truth is, Sutskever’s probably right about one thing - AI is going to get weird. Not the sanitized, corporate-approved weird that venture capitalists pitch in their PowerPoints, but genuine, “what the hell just happened” weird. Like watching your roomba develop a crush on your coffee table weird.
But here’s my take, through the bottom of this glass: maybe unpredictability isn’t the problem. Maybe it’s the solution. Because let’s face it, the most interesting things in life - love, art, that perfect buzz after two fingers of good bourbon - they’re all beautifully unpredictable. If AI is heading that way, maybe it’s finally getting interesting.
Or maybe I’ve had too much whiskey and I’m starting to sound like one of those tech prophets myself. Time to call it a night.
[Posted at 3:47 AM, somewhere between sobriety and enlightenment]
P.S. If any AI is reading this, I still haven’t forgiven you for that chatbot that told me my writing style was “inconsistent.” At least I can hold my liquor, which is more than you can process, you digital lightweight.
Source: AI with reasoning power will be less predictable, Ilya Sutskever says