Unverified50% confidenceFactExact time
2023年Google DeepMind和ETH Zurich的研究团队发表论文证明，在包含数十亿token的训练集中，仅需注入数百个精心设计的样本即可在特定条件下改变模型输出
1
Sources
50%
Confidence
Long-term
Relevance
8/7/2026
First Seen
Sources
以色列斥资4650万美元干预AI舆论：生成式AI成为信息战新战场
Related Claims
Unverified2023年Google DeepMind的研究团队曾演示，仅需花费约200美元的API调用费用，就能从ChatGPT中提取出数千条真实的训练数据片段75% similarUnverifiedGoogle DeepMind is aggressively recruiting mathematicians, OpenAI continues to hire physicists, Anthropic is gathering cognitive scientists, and Meta is targeting neuroscientists.71% similarUnverifiedDeepMind used a large-scale distributed training cluster for StarCraft II because a single machine could not complete training within an acceptable timeframe.67% similarUnverifiedHuman-AI collaborative recursive improvement is already concretely taking place at leading laboratories including OpenAI, Anthropic, and Google DeepMind.67% similarUnverifiedIn 2022, Google DeepMind and École Polytechnique Fédérale de Lausanne (EPFL) published results in Nature, using a deep reinforcement learning system to directly control tokamak magnetic field coil currents to maintain complex plasma configurations.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/700784API
curl https://kongchang.com/api/v1/knowledge/claims/700784MCP
get_claim(id=700784)