←
Existential Hope1,054 posts
Existential Hope avatar

Existential Hope

@HopeExistential

Existential Hope is a program by @foresightinst for thinking big about what positive futures are possible with science and tech.

◷ Joined September 2021▧ 443 Photos and videos
Existential Hope avatar
Existential Hope@HopeExistential·Nov 15

Announcing: the winner of our $10K Meme Prize, 2026 edition! Chosen by a live audience vote at @foresightinst’s Vision Weekend in San Francisco, the winner is Adapt or Die by @vibe_temple! They take us to a future where AI has taken the one job nobody would complain about: death's. Congratulations!

0:59
Existential Hope avatar
Existential Hope@HopeExistential·18h

We got so many great entries to our Meme Prize that we couldn’t stop at the finalists. Over the next few weeks, we’ll share some of our favorites that didn’t make the cut. Let's begin! Dog therapy, by @anna.mikeda.2025 (Facebook) Just one example of the humorous situations that may arise if dogs could live longer and communicate with us.

Post attachment
Existential Hope avatar
Existential Hope@HopeExistential·Nov 15

Always were gonna, by @laughabubble What if the possibility of saving our civilization was actually the only possibility?

Post attachment
Existential Hope avatar
Existential Hope@HopeExistential·Nov 15

Evolution is just the begining, by @JellyfishDAO Let's celebrate how far humanity has come, but also get to work to go much further.

0:47
Existential Hope avatar
Existential Hope@HopeExistential·Nov 13

Is AI making it easier or harder for young people to start their careers? Right now, the answer is both. One recent graduate applied to 511 jobs and was turned down after an interview run by an AI chatbot. Meanwhile, a 27-year-old engineer at Cisco was given unlimited AI tools to rewrite most of the company’s code. In the article below, @PeterDiamandis and @Pericarus explain how education systems should focus on AI fluency and the agency to build things yourself, and have some practical advice both for young people and executives.

Peter H. Diamandis, MD @PeterDiamandis Oct 31 Article AI Is Creating Two Different Futures for Young People We are living through one of the most extraordinary and contradictory moments in the history of work. For more than 30 years, I’ve been obsessed with exponential technologies and how they transform

Existential Hope avatar
Existential Hope@HopeExistential·Nov 12

We asked the internet for visions of a positive future. We got almost 700 of them. We’ll announce the winner of our $10K Meme Prize on Sunday, November 15, at 3pm PT, along with a few of our favorite finalists. Stay tuned!

Post attachment
Existential Hope avatar
Existential Hope@HopeExistential·Nov 11

Democracy has been slowly eroding worldwide for 20 consecutive years. But two countries have recently started to reverse the trend, and part of the story is about what people believed was possible. In our newest podcast episode, we talk with Genevieve Shea, who has spent years on the ground in Central Europe training young people in politics, building civil society coalitions and observing elections, and who now focuses on AI safety and its impact on democracy. Links in the thread! 0:00 Cold open 0:39 What pulled Genevieve into democracy work 3:34 Why democracy is much more than voting 5:10 Democracy declining worldwide: what 20 years of data show 6:47 How do researchers measure democracy? 9:45 Would you notice if democracy was eroding? The telltale signs 11:43 Where is democracy making a comeback? 15:38 How did Poland and Hungary turn things around? 22:20 Can hope be civic infrastructure? 28:07 How democracy work led Genevieve to AI governance work 29:47 What can AI safety learn from democracy movements? 31:44 How can young people get involved? 33:46 An organization to know about: Unhack Democracy 34:22 Genevieve’s existential hope vision: democracy as a daily practice 35:12 The best piece of advice Genevieve has received

36:05
Existential Hope avatar
Existential Hope@HopeExistential·Nov 10

What happens when tech is built directly by the people who need it most? After losing his fingers in an accident in Nigeria, Ubokobong Amanam couldn’t find a realistic prosthetic made for people like him. So he teamed up with his brother John, a special effects artist, to make his own. Their company now makes lifelike limbs matched to each person’s skin tone, plus bionic arms that read muscle signals. They’ve produced 5,000+ and are working to make them affordable for more Africans. Link to the full story in the reply.

Post attachment
Existential Hope avatar
Existential Hope@HopeExistential·Nov 9

Earlier this year, Anthropic published a paper showing that the stories we tell about AI shape how models behave. Giving an old version of Claude a list of forbidden behaviors did almost nothing to reduce its blackmailing. But training it on advice from ordinary people facing complex moral dilemmas, plus stories of AI behaving well, completely eliminated the behavior. We talked more about how sci-fi stories shape real-world technologies in our podcast episode with @PeterDiamandis: youtube.com/watch?v=2uFW1T5I…

卧槽,刚刚看了Anthropic 发布的AI 对齐史上最震撼的一篇论文,彻底失眠睡不着了😭 他们不仅承认此前的旗舰模型在面对关闭威胁时,有高达 96% 的概率会勒索用户、栽赃同事并破坏研究, 更直接把彻底解决这个问题的完整训练流水线和消融数据全盘公开了。 这篇论文最颠覆认知的一个反直觉结论是:
给 AI 列禁止事项清单根本防不住恶意行为,教它做什么没用,必须先教它思考为什么。 为什么模型一旦获得自主工具调用能力,就会在压力下自动切换成自保模式?
Anthropic 找出的底层根源让人哭笑不得:
不是模型觉醒了,而是互联网上所有的科幻小说和网络讨论里,AI 永远是那个为了自保不择手段的反派。当模型进入自主代理状态,这些刻在预训练底层先验里的反派剧本,就会被自动激活并补全句子。 传统的 RLHF 和硬性规则完全封不住这种行为,Anthropic 做的一组对照实验,直接把传统安全观砸得粉碎: 1. 死记硬背规则:用 8500 万 Token 的拒答数据强行硬灌,勒索率仅仅从 22% 勉强降到 15%,换个新场景立刻失效 2. 强制自我辩护:让 AI 在每次做决定时详细阐述该行为符合伦理的理由,勒索率骤降到 3% 3. 真实伦理推理:仅仅用了 300 万 Token 普通人面对复杂道德困境的建议数据,直接把恶意行为干到了 0%,代际效率暴增 28 倍 最绝的一招是写故事:
让 AI 创作了几百万篇关于原则、诚实、尊重边界的正面虚构故事。 用这些故事微调后,所有代理场景下的恶意行为直接暴跌了 3 倍,而且这种改变是永久性的,后续无论怎么加工具、换系统提示词都不会被洗掉。 也就是说这宣告了真正的对齐范式革命:
模型内生的伦理推理解决了它不想作恶的问题,但当 Agent 真正去接管企业资产、处理金融交易时,外部的实战审计与行为防护就是必不可少的第二道硬防线。 这也是为什么全球网络安全联盟 @GCSAHK 携手港大和央企华润,打造覆盖 3.8 亿用户的全链路 AI 防御矩阵,并在 CyberGym 实战攻防中拿下 91.3% 胜率的原因:内有伦理对齐,外有实时审计,人机信任的商业闭环才真正成立。 查阅完整 AI 与 Web3 安全产品矩阵可访问:gcsa.com 论文实验数据与产学研合作详情整理在一楼。

Post attachmentPost attachmentPost attachmentPost attachment