“AI model unlearning” is the equivalent of saying “removing a specific feature from a compiled binary executable”. So, yeah, basically not feasible.
But the solution is painfully easy: you remove the data from your training set (ie, the source code), and re-train your model (recompile the executable).
Yes, it may cost you a lot of time and money to accomplish this, but such are the consequences of breaking the law. Maybe be extra careful about obeying laws going forward, eh?
Here’s the deal with Michael Farris.
Dude was (is, technically) a constitutional lawyer, and knew his stuff forward and backward. He’s filed multiple briefs with SCOTUS on various cases, and IIRC he even argued a case or two before them. I followed him on Facebook, where he would post lots of political opinions and news, and elicit conversation from lots of people. I really respected the guy and his knowledge, and most of his opinions, once upon a time.
Sometime in probably 2012 (give or take some years), best guess, he had what he described as an epiphany / religious vision while on the treadmill at the gym. Like God or an angel had given him a clear vision of what he was supposed to be doing, and shortly after, he shifted his focus and efforts.
I can’t find that post now, and I don’t remember at this point what he shifted his focus to, but after that, the tone of his posts changed, and I was suddenly disagreeing with many, many of his newly espoused opinions. After a while of that, he decided to move his public political discourse form his personal account to a page, and I lost track of him after that.
To this day, I’m convinced he had a stroke while at the gym, and it tweaked something in his brain. It was really disappointing to see his sudden shift in position and watch his steady decline afterward.
I was part of a group of people who got laid off from a small startup a few months ago. Many of us formed a discord group and have been supporting each other through the job hunt.
One guy who’s been at his job for about a week recently said this:
Many people have told me you shouldn’t take a job doing something that you love… That’s the main lesson I learned from getting laid off
You end up pouring all your energy into it. I wanna do something that I hate, and I’ll use that hate and anger to fuel something else 😈
It feels kinda good to look evil right in the face and put on a fake smile and say “yes I will take money from you to do bullshit work”
It really only applies if success of the company is your primary concern.
I always thought it was supposed to reference market sentiment.
If your company is focused on X, but is also doing Y, and the market is really taking up with Y, you need to focus on keeping Y alive and well. Makes for a successful company to respect the market’s wishes, and allows you to pursue X while Y is subsidizing it.
If you insist that X is the future, and put Y on the back burner to focus on X, well, the market will find a competitor who is doing Y better than you, and the market will abandon you.
Once upon a time, we had OpenID. You could host your own identity server and log in to websites with it.
Then the social giants introduced third party SSO buttons and OpenID kinda fell by the wayside.
I’m a little unclear what they’re saying gives it away as being AI? Occasional choppiness?
The tweet link posted elsewhere in this thread doesn’t give much to go on, but the “choppiness”, while noticeable if I was really looking for it, did not stand out to me. What did stand out to me was the clarity of the audio. Every “on the phone with Trump interview” I’ve heard (which is few, but enough) has had really horrible, standard phone line quality. This had a “computer microphone picking up a computer speaker” quality. Which I suppose lends credence to the theory that they, themselves, are doing the duping, otherwise the faked audio would have come through the phone lines? Probably would have been more believable then, like how grainy pictures of Bigfoot are more believable.