Posted on:
I actually like X née Twitter. It's useful for keeping up with AI-related stuff. When I see people post on Twitter itself about Twitter "being toxic", it feels like they're complaining about a party guest they specifically invited over getting drunk and throwing up on the furniture. Maybe just don't invite that person over again? Maybe use the block/unfollow controls that are not hard to find in the user interface?
This is not a post about Twitter. Rather, it's about living here:
A lot of the AI-heavy Twitter contributors that I follow seem to live in the very early part of the Rogers curve. I don't blame them; it's fun and exciting there. I mean, they can't fully live there unless they also live in some kind of isolation pod with drones delivering their every bodily and physical need. But they publish like people who mostly live in that innovator/early-adoper part of the Rogers Curve.
And when these folks talk about how AI is going to impact the world, or some part of it like the job market, it's LOL funny the kind of crazy stuff they come up with.
Here's a good antidote to that:
This fallacy explains so much AI hype. The barriers to doing real-world tasks efficiently are rarely computational — they are usually physical, economic, social and political. So you could make the computational part 10x faster yet make the overall system only 1% more efficient. https://t.co/TvVMZEWg3c
— Arvind Narayanan (@random_walker) November 25, 2024
The rest of the system (and there is always a "rest of the system" -- always other parts of the system) may not live in the same part of the Rogers Curve as the part innovators/early adopters are focused on does. In fact, it's a safe bet that it doesn't. If it did, they'd be focused on it. They'd be fascinated with it.
Evaluations are all we need by Rohit Krishnan does talk about evaluating LLM output, but it's really a broader, higher-level reflection on what an incredible mess of complexity any actual "job" is. Parts of some jobs easily yield way to the efficiency gains of cutting-edge, unreliably-but-amazing innovations. But those parts of jobs are just parts of a bigger, messier connected system. And the job is not the parts that easily yield their inefficiency to innovation; the job is the whole system. Rohit's article is another helpful counterbalance to the "OMG look how many jobs AI just nuked with the latest releast of XYZ yesterday!!!" kind of stuff that's drawing a lot of heat nowadays.
Tagged with:
More posts: