Hacker News new | ask | show | jobs
by june_twenty 1122 days ago
Altman seems inexperienced for this type of role. I guess he didn't really expect OpenAI to be the leader of the AI race.

Hopefully strong competition comes along soon.

4 comments

It's easy to say that, but who would make a better candidate? It's a very weird and unique situation he/they are in. He just looks young, he's been around the block a couple times.
Agreed. I hope there is competition so there's not just one person, alone.
Do you understand that NN is not like rows in database. The machine can't unlearn what it has learned. You can't make AI unlearn something. They are going to ingest large amounts of data and trying to stop it is going to be nothing more but layering endless filter in front of it. Unfortunately or fortunately, it can be prompt broken. So if they try to abide and still fail, they will be fined enormously. See Meta & Google for example. This is about leverage. OpenAI has leverage and I'm happy to see a leader have balls. Let EU go back into the stone age if that's their wish.
So lets just change laws to accommodate an unethical data model from a corporation pushing techno feudalism?
> The machine can't unlearn what it has learned. You can't make AI unlearn something

This is not even remotely true: From https://en.m.wikipedia.org/wiki/Catastrophic_interference:

> Catastrophic interference, also known as catastrophic forgetting, is the tendency of an artificial neural network to abruptly and drastically forget previously learned information upon learning new information.

OpenAI could roll back to a previous version of their model.

Anyway, I'm not sure if you intended that as a reply to me because it's not relevant.

How? Do you think they have a previous model that's missing one single data point for anything that needs to be forgotten? The only way this would work would be to filter out whatever the EU demands then retrain the model without that data. So every time they get a request, they will have to retrain the model. Scale that.
I don't care how they scale it. They obviously have versions of the model and can go back to a previous version.
Strong competition means FAANG wins.
and hopefully unlike openAI it really is open this time.