正在加载视频...
视频加载失败
Jacob Coxon’s next interview on BBC (the ex-Anthropic+OpenAI researcher who resigned). "If we don't slow down at the current rate of progress, there is a strong chance that we could all die in the immediate future. It's not at all an exaggeration. And many of the people who are... show more
19 条评论
The fact that he's being interviewed by the biggest conservative manipulator there is in the BBC makes me put in doubt those claims and think of yet another manipulation scheme.
When you're genuinely 10%+ convinced AGI will kill everyone, you don't keep shipping it. You either leave (Jacob did) or the risk isn't really 10%.
isnt he in a busiless iwhich sells AI safety services now? so...
Unclear to me how 'we might all die' and 'ship it anyway' coexist in the same head. maybe that's the interpretability problem we should be studying
The "every day" part is the hardest to ignore.
A forecast of extinction is easy to dismiss until it comes from people who spend their days measuring the failure modes. The strange part is how quickly normal work can make an extreme risk feel routine.
Can we just YOLO into RSI and get it over with? That's the only thing that's going to shut these f***tards up
That's some serious food for thought. It's de
@hilbertspaess Jacob is apart of a cult who lives in shared housing with altruistic evangelists .. insanity they need to be taken to insane asylum.. they have no proof ai will do this.. based on sci fi theories
AI巨头们要求放缓的本质是AI目前发展存在的两个瓶颈： 第一是可以自我迭代的AI系统会将任何主观对齐方式给清除掉，而客观对齐方式又没有建立，会出现无对齐的AI系统，这就是ai巨头们口中所说的安全问题，但这个问题更多的是基于技术路径问题可以基于技术的方式去解决。这里还会涉及一个法律的来源及形成机制问题，该问题可以在系统内设置一个民主的法律形成与清除机制，具体方法在本人五态循环宇宙论的第四层内容里。 第二是闭源ai模型都是采用的集中式算力，当AI智能进入自我迭代状态时，大众对ai的使用会使对算力的需求急剧膨胀，会让闭源AI系统陷入崩溃。这是基于现实算力受限不得不采用放缓给自己补充算力或将集中式算力改为分布式算力，由闭源模型改为开源模型。解决方案是让每一个使用AI系统的人都具有自己的基本算力，让占ai系统80%的普通功能由使用者自己的算力运行，让占ai系统20%的高端功能由ai巨头们的集中式算力提供。 总结：以上两个问题都是AI发展过程中需要正面面对及解决的问题，加快发展去解决的问题，而不是通过消极放缓就可以解决的问题。更何况放缓只会让更多的AI系统到达自我迭代的分界线，让更多的ai系统同时面对以上两个问题，并不能解决以上两个问题，或者说反而进一步放大了AI无主观对齐但客观对齐又未完全建立时的风险。
Stop giving him a platform, pls 🙈
🧐
Incredible quote—existential AI safety warnings from high-level insiders drive massive engagement and urgent discussions!
When the engineers training frontier models publicly admit to a >10% extinction risk, it’s not marketing hype—it’s an open containment alarm.
That warning is chilling. Along with @AdamJohnstph, you're at the top of my feed—this interview is one to watch.
Full video on BBC.
Doom timelines sell headlines. Operators ship with rate limits, audit logs, and a named owner for every agent that touches money or PII.
Dario's own essay puts a number on it: a swarm capable of a persistent botnet in 6-12 months, hundreds of billions in damage. Coxon's "immediate future" and the CEO's timeline are basically the same clock.
model swap is easy. the workflow is what they keep
