Using AI today can sometimes feel like the first time driving down the highway. Or the first time snowboarding down a black diamond slope.
It was a packed week again. Lots of calls to prep for panels and gigs in the coming weeks and months. And clarity on a project I am quite excited about in 2027. To find that elusive spine of AI risk management and assurance. Also did a podcast, and an AI for Good panel where the panel itself was shorter than the time we spent waiting around because the prior panel overran.
Around me, the financial markets fell hard. The AI trade seems to be losing its lustre, especially after the US mid-terms. I thought this was a good thing. While I’m not in the camp thinking that we need to pace because of existential threats, I do think that slowing down the hype and bubble around AI can only be a good thing. I suspect many who punted on AI may be feeling some loss of control.
And I thought of two experiences that taught me what loss of control actually feels like.
The first was learning to drive when I was 20. When I first drove on a highway in Toronto, Canada. The second was finally getting the confidence to try snowboarding down a black diamond slope somewhere in Montreal, Canada.
Both felt similar the first time. But became quite different over the years.
Highway and slope
The highway in Toronto, that first time. A dim winter day. My instructor, a naturalised migrant from Hong Kong, told me to turn onto the ramp. And then I was on it, and the other cars did not slow down for me. I held the wheel tightly. And I was both tense and thrilled. My hands ached by the time I exited the highway. I remember how slow that grey road was on the way home.
After getting my licence in Canada, that thrill disappeared quite quickly. On most road trips, I preferred to just get the drive over with. Rather than being behind the wheel, I would rather nap in the back.
The slope was outside Montreal. I forget the name of the ski resort. I just remember that it was exceptionally cold. And that I stood at the top of the black diamond slope much longer than I needed to. And slowly inching down by ploughing down. Before getting some confidence, and tipping in. Snow, the edge of the board, the scrape of it, no space for a single other thought. I remember how flat the bottom of the slope felt.
After that run, some of us got addicted to black diamonds. That pensive feeling at the top never left. But neither did that thrill and rush. If anything, it stayed just as fun even as we gained more control.
Both felt the same the first time. The same loss of control. One of them slowly turned ordinary. The other never did.
When I speak to people using AI these days, some seem to be on a highway. Others on a black diamond.
And I feel like this has something to do with learning these days too. Someone told me this week that she was afraid of getting stupid from using too much AI. That if you hand over too much of your thinking, you get worse at it.
I told her I feel the reverse. If anything my head feels like it is exploding. Every week, I have more ideas than I have mental space for. More things I want to build than time to build them in. And learning, more and more with AI, about how many things I never knew I never knew.
I think it comes down to whether we are on the highway, or the slope. Hand over the highway, the dull and repeatable thinking, and that is just sensible. Hand over the slope, the hard stretch that was making you sharper, and yes, you get duller.
Enterprise and frontier AI
The same applies to the worlds of enterprise AI and frontier AI.
I keep hearing people explain that enterprises need to change their workflows to get value from AI. Which is true. It was also true last year, and the year before, and it will be true next year. It has been true for AI, digitalisation, computerisation, and maybe any technological change. This is the highway. Sensible, necessary, and said so many times that the words barely register any more to most.
In the other world, a research group got a frontier model to read another AI model’s weights, the raw numbers inside it, and rewrite a tiny slice of them directly, with no retraining at all. A small percentage of the weights changed. The performance of the AI model jumped. One AI model just changing the weights of another AI model. For someone who grew up training and tuning AI models like me, this is a paradigm shift. The first thought in my head is not “not again”, but “how the hell did it do this”! This is the black diamond. Nobody is quite sure where the edge even is.
Both are AI. The first is reminding you to tidy up your processes. The other is an AI model being able to edit the inside of another model. We talk about them as if they are the same subject. They are barely in the same world.
A boring but interesting life
And ironically I want both.
I want a boring life. I’ve enough. So I choose not to sacrifice time and health chasing things I’ve no interest in just because people do it. Or saying things I do not believe in. LLMs today are also already capable enough for me. I could be quite happy even if the pace of new models appearing slowed down 2-3 times. The profits and losses of the AI companies matter little to me.
But I also want an interesting life. I want the black diamond. I want the pause at the top and the few seconds of not being quite in charge. I want my head too full of ideas that each day feels like an adventure.
What’s strange is that I am not sure these two wants of mine have ever sat this close together before. I want to make AI boring and predictable with AI governance and risk management. But I also want it to stretch the limits of my imagination and capabilities. Like the management consultant, I build highways for a living. But I also go looking for black diamonds.
We live in a paradox.
#AI #AIFluency #AIGovernance #AILiteracy #Reflections #AIRiskManagement