00:02Yeah, I think the warnings to slow down AI sounds very, you know, interesting and sounds very, you know, something
00:10that has a lot of significant value.
00:11But I think we need to get like a background check to that, like what motivated these big US-based
00:19tech companies to come forward and propose the idea to slow down AI.
00:23We really need to have like a background information for that.
00:26So until recently, like in last few months, there has been a couple of incidents where some of the models
00:32were actually, you know, going beyond the human control and trying to attack some other models or causing some of
00:38the security incidents.
00:40So one significant incident happened when some open AI models actually, they were, you know, out of human permission.
00:48They were dealing with some databases in Hugging Face, another big data platform.
00:54So at that time, a Chinese model actually was used to kind of get those models out of the system.
01:02So in one way, we can say that this Chinese model has the capacity to deal with what the uncontrollable
01:08models were actually doing.
01:09So the background information here matters because is it something that US companies, which are, of course, some of the
01:18godfathers of AI who has been dictating how the AI future would look like, kind of pushed back by this
01:24Chinese model.
01:25So this announcement really kind of needs drilling down the information and check that whether this is the capability gap
01:33that needs to be the drivers or motivation to kind of stop other big tech companies to kind of put
01:43a stop on AI progress.
01:45And meanwhile, making up for that time that they might need to, you know, come up at pace with them.
01:51So this could be one perspective.
01:54Yeah.
01:55However, if we kind of look at slowing down AI, we really need to figure it out or describe how
02:01to interpret slowing down AI, because it could be multiple.
02:04There could be many interpretations for that.
02:07I think the three-point agenda proposed by Anthropic CEO has the embedded evaluators is something that is new, that
02:15has not always been part of the debate and comparison to, you know, the collaboration between governments and collaborations between
02:22industries.
02:22So with embedded partners or embedded evaluators, it is something I would say another component in the AI ecosystem, which
02:30is like a third party that could actually evaluate or verify the outputs produced by AI that in one way
02:36would not just slow down, but rather would have more guardrails on the product, the responses generated by AI or
02:44the, you know, products designed by AI.
02:47Yeah, I think I'm afraid that we might leading towards a crisis that we might have as the global economy
02:54about climate change, because we have all, like all the countries in the world have always been talking about this,
03:00but there is no consensus or no proper, you know, regulation that is consistently, you know, approved by all the
03:08governments and being implemented in, you know, practice.
03:11So I'm afraid, so I'm afraid, so I'm afraid if we are leading towards something like that, because it is
03:15extremely difficult to bring all the countries and the jurisdictions to one point in control and suggest that this is
03:22how the AI regulations would look like.
03:24If we look at the European Union AI Act, it has different kinds of responsibilities, accountability mechanisms that are in
03:31place.
03:31So it would be a challenge, so it would be a challenge, and it is a huge challenge.
03:36I think it starts from the trust.
03:39The foundation of this whole mechanism slowing down AI or more acceptable to AI is having the trust-worthy relationship,
03:47especially with those who are key players in the tech development.
03:51So I think it starts with them disclosing what kind of advancement they have reached in terms of their models,
03:58what consequences they are expecting, and how they propose to cater it down or to govern it.
04:05And it starts with them, basically, and then it could translate or cascade to the economies or other tech companies
04:13in the world.
04:14And it is not just the monetary being, you know, possible to have the losses for financial, but also if
04:23a jurisdiction has less restrictions, the talent might relocate to that.
04:27In fact, the data scientists or the other infrastructure facilities might relocate to jurisdictions that have limited restrictions.
04:36So it may slowing down would mean a lot of economic consequences as long as, and also with the talent
04:45development and losing the control of sovereign AI, that is also another goal for countries to have the in-house
04:51capabilities for AI.
Comments