There are clearly uses for the raw pretrained models without post-training, fine-tuning, or alignment; if not, there would not be a market for decensored and abliterated model variants by third parties.

If reputable model publishers could instead publish their models in verifiable parts: pretrained weights, fine-tuning/alignment adapters, and a combined intended-use model - this would reduce the attractiveness of malicious fine-tune publishers. I'm not holding my breath for that though, quite the opposite. I expect the economic and power incentives will encourage the big training players to push for ever-higher but nebulous safety regulations until everyone else is choked out except for them.

Until training costs come down, and perhaps even after that, all AI should be treated as untrusted rogue agents, constrained by default regardless of source. That's just the sensible thing to do, but the economics will have something to say about that.

> The focus on deepfakes in a vacuum is actually missing the forest for the trees, with the ‘forest’ being the fact that we now rely on five massive social media platforms as the arbiters.

I think the way legislators have come to understand media - at least by the way laws are being written - is that the collection, distribution, and filtering must happen on a single platform. So there's this attempt to force platforms to be the arbiters of truth.

But this is dangerous in several different ways:

1. It is impossible to be an arbiter of truth, it's a fundamental problem of epistemology that cannot be solved without perfect knowledge of the "real world", and so the platforms must necessarily be allowed to fail to some, or a large degree.

2. Having been given this impossible responsibility, the platforms are going to disallow/punish based on controversy, because their real problem isn't breaking the law, it's others bringing attention to where they're breaking the law.

3. In the meantime the platforms might offer some token resistance to these laws, but ultimately they will not say no to being handed the one tool that gives them enormous political and economic power, power which will protect them if the law comes after them.

These three points will keep reinforcing themselves to lock us into the information dystopia we're in today.

I don't think it's possible to preserve 1. freedom of speech, 2. protection from mis/disinformation unless the different layers could be forcibly separated. My current crackpot idea is that the distribution and filtering layers have to be broken up in a way that both forces information to flow freely, and allow consumers to protect themselves by consciously selecting and filtering their consumption in meaningful ways. Filtering providers cannot be given the power to control information supply for other filtering providers, which allows the market to abandon untrustworthy providers at will, and also gives the authorities the degree of power required to find bad actors and dispense meaningful punishments.

[0] https://www.gov.uk/home-education

What if creating easily-fusable isotopes is the most efficient way to store "renewable" energy? Fusion is the most concentrated release of power humans are able to muster. What's to say such concentrated power will never be required for any purpose?

Statistically speaking one could be fairly confident, but as a rule it is wrong due to the presence of compound surnames.

https://en.wikipedia.org/wiki/Chinese_compound_surname