Starlight / Compass / Potential T6
Potential T6 193 records
A high ground already under attack
English titles and summaries are available for 347 of 1154 catalogue records. Untranslated cards retain clearly marked Chinese originals; detailed records remain in Chinese.
最小基因组合成细胞
把生命需要的基因删到不能再删，剩下的就是生命的底层操作系统
AI可纠正性·防止规避关断
让足够强的AI系统始终愿意被人类叫停,而不是想办法规避
空间太阳能无线能量传输
把太阳能电站搬上轨道、把电波打回地面——如果成了，人类第一次拥有不分昼夜的清洁基荷电。
区域 firm-power 光伏储能平价
让'不分昼夜阴晴、比烧煤烧气还便宜'覆盖全球电网,不只是最佳光照区
Aligning recommender systems with human well-being
Distribution algorithms in effect govern the flow of civilization’s attention, yet their objectives are systematically misaligned with people’s long-term well-being.
Electrolyzer scarce-metal substitution: iridium-free oxygen evolution
Global annual iridium production is about 7 tonnes, so terawatt-scale green hydrogen does not add up on the materials ledger. Acid-stable, iridium-free catalysts are its physical bottleneck.
AI tutors: universalizing education’s two-sigma effect
One-to-one tutoring can raise ordinary students by two standard deviations, but has always been too expensive to universalize. AI makes “a private tutor for every child” economically feasible for the first time.
深度学习泛化理论之谜
让过参数深网络的泛化从事后解释变成可证伪预测。
Mechanisms for distributing the automation dividend
When AI decouples output from labor, how should the increment be shared among all members? This is the largest distributional question of the intelligent age; there are only scattered experiments and no mature mechanism.
Formalized mathematics
Move mathematics into machine-verifiable systems: a foundation for trust and for AI mathematics.
Energy per bit: the network energy-efficiency wall
The physical cost of moving information is approaching the limits of existing devices, with no path to an order-of-magnitude reduction—an invisible ceiling on exploding AI-era traffic that is nearing uncharted territory.
Provably safe AI: formal assurance
Use mathematical proof rather than empirical testing to constrain AI behavior in critical infrastructure, upgrading “probably safe” to “provably safe.”
AI自主假设-实验闭环
让AI自己提问题、设计实验、写论文·科研生产函数改写
巨型戴森群碰撞级联防治
让数百万独立收集器不因轨道级联碰撞崩解成一圈太空垃圾
空间太阳能微波无线传能
让太空里全天候不间断的阳光变成地面可用的电
液氢储运材料抗氢脆规模化
让金属容器在液氢环境下不再悄悄变脆开裂
脂质纳米颗粒器官选择性投递
mRNA疫苗的LNP技术天生偏爱肝脏，让它学会去别的器官是下一关
HALEU燃料供应链瓶颈
先进反应堆没有燃料就是一堆钢铁
直接空气捕获能耗成本墙
从空气里抓碳的技术能跑通,但每吨成本还比目标价贵4到10倍
Proton decay
The direct experimental window on grand unification.
Fast-breeder reactors: closed fuel cycle
Raise uranium utilization from under 1% to about 60%, expanding fission’s energy ceiling by two orders of magnitude. The technology has been demonstrated for seventy years, but the closed loop has never been completed.
Interstellar exploration propulsion: reaching nearby stars
Use laser beams to push light sails, accelerating probes to a few percent of light speed and reaching nearby stars within a human lifetime.
Verifiable continual learning under limited resources
Let AI continually acquire new capabilities while measurably retaining old capabilities and safety boundaries.
冯·诺依曼探测器·星际自复制机群
能自我复制并携带全套制造能力的探测器,理论上能让文明指数级扩散到整个星系——目前只存在于纸面上。
1型糖尿病功能性治愈 · 让身体重新自己造胰岛素
不是控制血糖,而是让免疫系统不再攻击胰岛细胞、让新胰岛细胞真正定居
电网构网型逆变器 · 让新能源自己撑住电网频率
没有旋转的发电机转子,新能源怎么给电网'定音'
多组学生物年龄时钟的因果验证
现在的衰老时钟只会算命,还不会开药方
基因驱动反制与生态风险闭环
把特定基因驱动的受控反制、种群恢复、生态恢复与跨境治理逐层验明。
跨机器人操作平台的基础策略
让一个操作策略在多种机械臂和环境之间迁移，而非每台机器人从头训练。
体内基因编辑的跨器官细胞定向递送
将编辑器安全送到肝外的指定细胞类型。
戴森群长期自愈运维网络
让数百万收集器在数十年到数百年尺度自主更换损耗节点、不靠地球指挥
端到端驾驶大模型·从感知直达决策
用一个神经网络学会开车,不再靠人写规则。
仿星器构型优化 · 无电流稳态聚变
不用感应电流也能稳态约束等离子体的磁笼设计
星际激光推帆地基阵列
造一座百吉瓦级地基激光阵列,十分钟内把克级光帆推到五分之一光速
Tau蛋白病理清除
阿尔茨海默症盯了三十年的淀粉样蛋白之外，还有一个同样致命的靶点
EUV光刻胶随机效应
光子太少,芯片图案开始"随机丢线"
Page 2 of 6 · 193 records · ordered by potential score where present
← Back to Compass overview · Open this page in Chinese