Will OpenAI, Google, or Anthropic Have a SOTA Model's Weights Hacked Before 2025?
โž•
Plus
18
แน€937
Jan 1
13%
chance

Needs to be verified by an article in a major publication

Get
แน€1,000
and
S1.00
Sort by:

Does "hacked" only mean "stolen" in this context, or are there other types of events that would count as hacking?

predicts NO

@NathanShowell What do you mean?

@JordanSchneider213c What kinds of actions performed on the SOTA model's weights would you classify as hacking?

Why is China tagged?

@zyc well...

predicts NO

@zyc Dario Amodei said in a Dwarkesh interview said the following:

One of our goals is that it costs more to attack Anthropic than it costs to just train your own model. It doesn't guarantee things because, of course you need the talent as well so you might still, but attacks have risks, the diplomatic costs, and they use up the very sparse resources that nation state actors might have in order to do the attacks.

We're not there yet by the way. But I think we are at a very high standard of security compared to the size of company that we are. If you look at security for most 150 person companies there's just no comparison. But could we resist if it was a state actor's top priority to steal our model weights? No. They would succeed.

seems like the likeliest culprit?