3 ms·
> how has generative AI had a positive impact on the average person? Generative AI, for the most part, is a technology in search of a problem. For now. The te
by Swizec 3y ago
> how has generative AI had a positive impact on the average person?
Generative AI, for the most part, is a technology in search of a problem. For now. The tech is in a “demos well, productizes poorly” stage. Once you try doing anything at scale and with proper evals of success, you see that real world performance is pretty poor still.
Yes yes we have lots of totally-not-cherry-picked papers where researchers achieved something fantastic. Then you look at the detail and it’s either “we ran this once because expensive” or “it achieved this great result almost 35% of the time so it’s state-of-the-art best-in-class”
> create a techno-utopia, and they may be sorely disappointed when others don't feel the same way
Every techno-utopia I’ve ever seen in movies, books, etc has always been a secretly dystopia. Looks nice and polished on the surface, but achieves this result through aggressive oppression and disenfranchisement of dissenting voices.
Star Trek perhaps is the only one that didn’t follow that pattern. And even then you have groups outside the federation who are not super happy with how the federation does things.
- phillipcarter 3y agoJust to offer a counter-point, we had measurable success with this tech solving a direct problem it's designed to solve for: https://www.honeycomb.io/blog/we-shipped-ai-product https://www.honeycomb.io/blog/we-shipped-ai-product My take is that the tech isn't in search of a problem, we're just early. It solves a lot of problems. The challenge is getting the right data pipelines in place to allow it to do that. The good news is that in many cases, it's a good thing to have that in the first place.
- Swizec 3y agoI agree! GenAI is great at solving lots of problems right now, but not all problems as many are hyping. And for many use-cases it's still too expensive/slow. I wasn't trying to say it's useless, I was getting at this sentence from your 2nd paragraph: "[did it work?] The short answer is... mostly." That "mostly" part is the cold reality many products have run into this past year. It's been the biggest problem in my experiments at least. The current tech requires a lot of babysitting.
- phillipcarter 3y agoI would say that "mostly" is an incredible achievement. Most feature work done in software produces zero to middling desired outcomes.