Well, as I mentioned, the 23 times more than the previous daily peak last August was not the only time that this happened. Actually, earlier this year, when Dr. Tsai Ing-wen, our president, visited the US, there was also a very comparable foreign cyber-attack slash interference that happened. The reason ...

And so, I’m cautiously optimistic that even though the cost has massively lowered for someone who is not versed in a local culture to imitate someone else persuasively under the target culture, we can still have the awareness and societal resilience to counter those operations.

And so, yeah, recently, I think there is a societal recognition, there’s some news about essentially using VPN to fake basically a threat to a certain opinion leader. But it turns out that someone threated themselves via simulated persona online and so on. And that is actually quite a big ...

Yeah. So, in terms of scamming, this has already happened, including voice cloning and messages and so on. Whether it is used in the election context or not depends on the countermeasures that we’re taking. Because after all, while the cost has lowered, it is not zero, it is not ...

This is the kind of supernumerous threats that I’m describing here.

But now what this generation of AI tool does is not to markedly increase the quality of such scams, but rather to massively lower the cost. That is to say, as long as you have one domain expert and have the expert training the model, then you get maybe 80% ...

As I mentioned, this is mostly about lowering the cost. Previously, if you want to do precision persuasion or scamming and so on, you need someone who is versed in the local culture to do that. So, you can’t always do spear phishing or to do very elaborate scamming calls ...

And so, yeah, I wouldn’t use the future tense. Of course, all the frontier like GPT and so on have safety measures, but it is actually, if you look at the latest model like Mistral or Zephyr and so on, it is already exceeding the free ChatGPT in terms of ...

On the other hand, the technologies I talked about, the voice cloning and so on, many of which are just, you know, in the wild. That is to say, anyone can download and use them for free. And there are also open models with like on Hugging Face and so ...

Well, I mean, if you mean AI just as in ChatGPT or Claude or Bard, of course, maybe none of them are explicitly a reseller of these frontier AI models, because these have safety measures that prevents them to be used for such purposes.

There’s no robocalls or fraud calls in the US?

So, the ones that you have met face to face and or have common contacts and added to your address book, these are humans. But everyone else that doesn’t have a blue tick, that is to say, an authenticated number and so on, can actually be assumed to be a ...

So, my point being, as long as we can easily say that, OK, these are the authenticated, FIDO compliant protocols of sending information, everything else can be safely assumed to be probably a bot. And that flips a default because previously default is that people who send you a message, ...

Very quickly, we will go to a norm where if it is a respected agency or institution in the government, in the public sector, they will just communicate with such easy to identify way to authenticate themselves. And no matter how many other messages you receive, that purports to be ...

So, the idea here is not to get caught up in the tracing, like contact tracing, when the virality is this much, but rather to make sure. For example, just a couple of weeks ago, we launched a 111 platform so that all the governmental agencies can very easily send ...

The reason why is that GenAI has massively lowered the cost to, for example, imitate members of specific social or cultural groups to falsely represent grassroots support astroturfing. But not only that, but also instead of spreading, they can also do direct message or very small groups and disseminate content ...

Yeah, so in terms of the actor behavior content distribution framework, ABCD framework, tracking is really at a downstream level, the distribution level. And we are moving upward to the actor and behavior level now. Our main work is not on the content and distribution level now.

So, that is one very concrete thing, is to adopt this zero-trust network architecture. There are many more, but I would just highlight this one.

So, by going to three different vendors in biometric, in device, and in behavior, we can actually detect the attack attempts that may have breached one or even two of the three factors, but still mitigate so that it does not laterally move, meaning that it doesn’t spread to other ...

On the other hand, for example, when I sign my official documents, the FidO app on my phone checks for my fingerprint and an edge detection software, CrowdStrike, checks my phone and CloudFlare analyzes what the phone is doing behavior-wise and so on.

Now, as for concrete measures, for example, we switched off passwords in my ministry and the two administrations for industry and cybersecurity, respectively, because we understand nowadays with generative AI, spear phishing, phishing attacks, scam, voice cloning, video cloning attacks, and so on, are so rampant and can be mounted ...

Of course, that didn’t work out for the attackers. The stock exchange actually raised that day instead of plummet, as they hoped. But really, cyber resilience against foreign interference has been our main work since the beginning of the ministry. Our slogan is ‘digital resilience for all.’

Certainly. So, for us, who have started just a little bit over a year ago, last August, the same day, the same hour that our website went online was also the hour that missiles flew over our head following Speaker Pelosi’s visit. And at that day, we suffered more than ...

Sure.

Yeah, and thank you for accommodating the time differences.

Pretty good.

Hi Tiffany.

Awesome. Thank you. Live long and prosper.

Yeah, of course. It’s public, actually. There’s a PDF about that so I will airdrop it to you.

Anyway, but for internal uses, it’s the AIEC, right? The evaluation center. But we cannot skip our evaluation and testing, so it’s mostly just internal testing.

So, after we taste our own champagne then we will know whether it’s ready or not for something like polis.tw which is a public facing service. But there’s strict restrictions hardware in the GenAI use in the public sector guideline against directly providing public inputs into a GenAI unless it ...

But with some fine-tuning… we will be able to power some sort of Talk to the City with a locally hosted… I mean if you’re willing to throw to your edits and you can run Falcon-180B which is guaranteed to be good enough for this but it’s more expensive. So, ...

Yes. We now finally has access to the repo, and Odie already has the local models deployed in google chat mode and this is not a secret. So, we will see if our local model is good enough to power the Talk to the City. Probably not?

Guide rails, not guard rails.

Alright.

… AI set got rails for a conversation.

We signed the twenty first century agreement with US. And the core demand of not just AmCham but the European Chamber is on this, not just AmCham but European chamber of commerce too. So, they want to have a voice even though they’re Mandarin may not be perfect, they may ...

But again, like for truly controversial regulations, it’s not in a consumable amount, so exactly like regulations I’ve done, right? That’s something that Romney has solved very deeply about so I would consider and be expert in this. But I think regulatory good announcement is certainly very institutionalized in Taiwan.

That was the core proposal of Join Plus, the presidential hackers on the team, that basically say that if we tally everything in a regulatory pre announcement, which is the Join platform, it’s almost always now at least thirty days. And I think we’re seeing more and more solution to ...

So… but some sort of governance institutions need to effect change in order for these bridges to be considered meaningful. Otherwise… Well, we still get something meaningful which is the stakeholder groups come to trust each other more which may be the great thing already, but the point being that ...

Indeed. After the bridges are made between the same building groups or clustering, if result in governance changes either from any level of governments on regulations or maybe municipal level changes if it’s a river.

Yeah, we’ve went through the whole agenda for not eating anything from Wendy.

That’s probably not in Taiwan but somewhere else more polarization, yeah? Why just Cities, huh? Rivers!

Oh, that’s our AA report.

It is very clear that none of us have an idea.

Well then, it is not that cohere, right?

This is a good first start, right? Something is well calibrated if it is clear, if it is in the original context instead of taking the speeches, the cluster side of context, it retains the nuances, the specifics within the initial speech, that all the diverse compositions are there instead ...

And there is another case where this is just in a synchronous meeting and there is a demo, right? People seeing the interactions with those clusters or even in silico transcripts that is acted out by performance on stage and so on. And that people voice their support or concern, ...

So, I would argue that in that scenario, the success is rather the amount of further calibration that people did to that cluster model. Instead of whether it is initially very well aligned. But that is a harder scientific problem.

So, with technology such as this that can be hosted or can be even just run. People can just download the model that is representing the cluster to their own machine. One can explore more conversations in their own context that doesn’t take anybody else’s time, and see how calibrated ...