|
|
|
|
|
|
by aphextron
1857 days ago
|
|
|
>Take this scenario: You’ve hiked Mt. Adams. Now you want to hike Mt. Fuji next fall, and you want to know what to do differently to prepare. Ah yes, that totally common scenario which I'm faced with all the time. I love this. It perfectly illustrates the peril we are in with the current state of AI research. That the author would choose this as a problem to solve shows exactly the socioeconomic class they come from, and how that influences the way they solve problems. It may seem like a trivial and meaningless example, but these subtle biases will creep their way into these systems and be amplified. And you can bet that this kind of work is the foundation for what will become the technology that eventually governs every facet of our lives once AGI is a thing. I, for one, am terrified of the implications that a bougie tech bro AI overlord entails. |
|
|
If that's your concern though, the good news is that in its purest form, machine learning tends to bend AWAY from this. You need large data sets to get good results, which means these projects tend to sample huge chunks of the general Internet, not just the isolated bubbles of SV types. Of course this still has limits, any data set has limits. You can only scrape data from the net if someone has posted that data in the first place, for example.
But in their initial form, a lot of these models are pretty diverse. That's why AI Dungeon had all kinds of "objectionable" content that kept getting the always-offended on their case: GPT-3 is just built off the general Internet, including a lot of weird, fucked up shit. The real problem is that inevitably someone complains, and they start hacking away at the ideal model to try to make it squeaky clean and ruin it in the process.
If you want to keep the tech from being perverted by "bougie tech bros," focus on the censorship. The models often start off pretty good.