But one could argue that Taiwanese democracy is…

Why is Japan the natural baseline as opposed to Korea or Poland?

I guess I’m wondering, what is the actual vector for delivering the information?

Help me understand the mechanism. So if I have this great talking point that I think is really going to resonate, do I just email it to a journalist?

And why is it so easy to get them to pick it up? Let’s suppose that I want them to pick it up.

You just care about getting the narratives into the public.

You see a lot of volume in terms of hitting forums, private groups online and facebook, basically as A/B testing to figure out what resonates. And then it’s quite easy to take that, continue to propagate it, perhaps on social media, trying to hit virality.

Okay, so to make sure I have that part clear, so that you’re saying that there’s likely use of generative AI crafting these narratives. And typically it starts off some of the evidence here again, is like some of the cultural contextual differences that you see kind of raw red ...

So there’s no nuance here that I’m missing. Should I be understanding the Taiwanese context?

And then there were a lot of advertisements for health supplements for the elderly, a lot of advertisements for, I think I was pretty sure it was 國民黨 (Kuomintang) because the demographics of people watching me as the american looking at this, this is like Fox News and appealing to ...

So, yes, it was a very long conversation about the crime rate amongst minors, I guess, teenagers. Teenagers and a lot of polemics. A lot of screaming, a lot of yelling, a lot of back and a lot of agreement between the people, but they were all very angry.

So this is a broadcast I was watching in the evening. You know, it looked to me just like classical punditry, like we would have in the United states. I guess my question is, am I to interpret this in the same way I would interpret watching, say, msnbc or ...

With respect to kind of in terms of the existing mainstream media and kind of ideology affiliated. Just so there’s me, as somebody who’s new to kind of understanding journalism in Taiwan, I’m wondering if, so I was watching this last night and I just wanted to make sure I’m ...

But I’m trying to.

Just on one thing you just said, and I apologize. This is going to be a naive question.

Even there. Do you see just an increased amount of volume there, even despite the fact that you’re only limited to the…

Machine learning and foreign. You’ve mentioned volume. One of the concerns we had about the volume argument was that there’s a bit of a pipeline issue. You really basically have facebook and you can only generate, you can generate as much content as you want, but it all has to ...

There’s like, there’s enough cultural divergence between different groups of people using.

Can I try to think of another example? Just make sure I understand. So like maybe the use of maybe the 臺 in taiwan.

So the contextual disambiguation, that’s failing.

Okay.

There’s an example when there’s a many-to-one relationship between a traditional character and a simplified character, such that when you’re, when you’re just google translating from simplified to traditional, I would hope google translate would capture that nuance, but um [laughter]

Yes.

Yes.

Big old complicated traditional characters.

Can you give an example?

I wonder why that’s clear what the evidence is.

Well, I have a question. You can tell me if you want to order it differently, but I’m wondering if I’d like to hear more about this. So you just asserted that LLMs are being used to generate paperwork.

Right. But

The liability is no longer there. If somebody is now just trying to give you information that’s going to sway your vote, there’s no financial liability.

No upper limit.

I mean. But at the same time, if somebody, so if you force facebook to build an infrastructure for stopping grandmothers from getting bitcoin scammed because they would be financially liable for scams.

Civil liability.

Once you have this law passed. If, for example, a political entity, a foreign one, they’re trying to use a similar tactic to influence an election, could the law then be applied to curtailing that?

Is it just as simple as thinking like, well, the killer app for the same way that people often say that the early adopter, internet supporters here, like the early adopter for the killer app for disinformation would be scammers, the government will follow suit. Is that how?

I’m curious, from a standpoint of disinformation, you immediately connected it to essentially scamming. Right? And I like that analogy myself. In other words, if you’re trying to understand, say, a foreign state actor interfering with an election, perhaps the best place to look is understanding how those crypto scams, ...

When banks are in America, we will never ask you for any information over the phone. Or the IRS does that. I think it’s us. The irs says that they will only contact you by mail, never by phone.

For example, with anti-vaccination. So there are cases of people getting vaccinated and then dying, for example.

And that makes a lot of sense because a lot of the bad blood in these types of arguments comes from the assumption that the other side has bad faith.

Because silo cuts context.

What does that mean? Confidence building measure? I’m not sure I understand.

Yeah, I can imagine that was very, there was much head shaking going on.

Aunt and uncles depend on who the parent is.

How do you combat that?

Still, it seems to me that even if you were to address some law that, for example, the content that not even a catholic pushback, going back to Madeleine’s earlier point, it would seem to me that if, despite the wording of the law, if it’s portrayed by various political ...

When the government was one of those.

You’re saying that this kind of credibility with the public made them a source of power even in the early days.

Very early during the tsunami.

Very strong in the national presence.

Why not use the RLHF interface instead? I mean, if OpenAI can hire anthropologists and doctors and lawyers and stuff to do fine tuning to provide feedback, why not a grassroots approach to that?