Vitalik Buterin pfp
Vitalik Buterin
@vitalik.eth
One way to look at the possible effects of deepfakes in politics is to look at "shallow fakes" that have been possible for a long time. For example, this screenshot purporting to be a "you're free to say whatever nasty stuff you want" whitelist for twitter. I actually have no idea if this is real or fake, but there definitely are some people who are treating it as real, imo without sufficient evidence. And yet, the effects of things like this are pretty bounded. This feels like a plausible outcome for deepfakes in politics. The smart people know not to trust a politician's statements without confirmation on official channels, and less smart people get tricked, as they do today with shallow fakes. If something fake goes viral, Community Notes can help on the margin in making it clear that it's fake. Most normies don't actually look at this stuff in real time, and so the effect ends up not too bad. This is the strongest case for not being too worried. The place where I am *more* worried about deepfakes is...
47 replies
697 recasts
2703 reactions

Vitalik Buterin pfp
Vitalik Buterin
@vitalik.eth
... financial scams. This is an area where many people have no existing defenses - their family members desperately calling them saying they really need a big pile of money quickly. Or the corporate equivalent of the same. This is the reason I wrote this article: https://vitalik.eth.limo/general/2024/02/09/securityquestions.html Though we really need a much stronger concerted education effort in this regard. Also much better authentication across the stack.
10 replies
44 recasts
320 reactions

Maya Labs pfp
Maya Labs
@mayaprotocol
Absolutely, the whole financial sector is going to be massively affected by deepfakes being used for fake IDs & documents, personality impersonations and can result in complete chaos in the corporate and banking sector. We have been working on this problem for some time now. It's important to understand that a "social factor", something like you proposed "security question" is ultimately required. Anti-Deepfake Technology like ZK-based provenance or AI detection alone cannot solve this.
0 reply
0 recast
0 reaction