科技大佬的X

// DECODING THE FUTURE, ONE TWEET AT A TIME //
8
推文总数
14
关注用户
2026-07-27 13:23
最后更新
Ilya Sutskever
是时候扩展 SSI 了：转发 @SSI Inc.：我们宣布与 NVIDIA 建立长期战略合作伙伴关系。 NVIDIA 正在对 SSI 进行大量投资，这将使我们在未来 12 个月内将计算能力提高 10 倍。我们的研究已经达到了值得扩展的程度，通过这种伙伴关系，我们将能够做到这一点。我们是
Time to scale that SSI: RT @SSI Inc.: We are announcing a long-term strategic partnership with NVIDIA. NVIDIA is making a substantial investment in SSI that will let us 10x our compute in the next 12 months. We reached the point where our research is worth scaling and with this partnership we will be able to. We are
Ilya Sutskever
我们宣布与 NVIDIA 建立长期战略合作伙伴关系。 NVIDIA 正在对 SSI 进行大量投资，这将使我们在未来 12 个月内将计算能力提高 10 倍。我们的研究已经达到了值得扩展的程度，通过这种伙伴关系，我们将能够做到这一点。我们是
We are announcing a long-term strategic partnership with NVIDIA. NVIDIA is making a substantial investment in SSI that will let us 10x our compute in the next 12 months. We reached the point where our research is worth scaling and with this partnership we will be able to. We are
Ilya Sutskever
🇺🇸🦅250！
🇺🇸🦅250!
Ilya Sutskever
Anthropic 没有退缩，这非常好，而且 OpenAI 采取了类似的立场也很重要。未来还会出现更多此类具有挑战性的情况，相关领导人必须挺身而出，
It’s extremely good that Anthropic has not backed down, and it’s siginficant that OpenAI has taken a similar stance. In the future, there will be much more challenging situations of this nature, and it will be critical for the relevant leaders to rise up to the occasion, for
Ilya Sutskever
我提出的一点没有被提及： - 扩展当前的事情将不断带来改进。特别是，它不会停滞。 - 但一些重要的东西将继续缺失。 转发 @Haider：以下是今天的 ilya sutskever 播客中最重要的几点： - 5-20 年内的超级智能 - 当前的扩展将严重停滞；我们回到真正的研究 - 超级智能 = 超快的持续学习者，未完成的预言 - 模型的泛化能力比
One point I made that didn’t come across: - Scaling the current thing will keep leading to improvements. In particular, it won’t stall. - But something important will continue to be missing. RT @Haider.: here are the most important points from today's ilya sutskever podcast: - superintelligence in 5-20 years - current scaling will stall hard; we're back to real research - superintelligence = super-fast continual learner, not finished oracle - models generalize 100x worse than
Ilya Sutskever
以下是今天 ilya sutskever 播客中最重要的几点： - 5-20 年内的超级智能 - 当前的扩展将严重停滞；我们回到真正的研究 - 超级智能 = 超快的持续学习者，未完成的预言 - 模型的泛化能力比
here are the most important points from today's ilya sutskever podcast: - superintelligence in 5-20 years - current scaling will stall hard; we're back to real research - superintelligence = super-fast continual learner, not finished oracle - models generalize 100x worse than
Ilya Sutskever
重要工作 转发 @Anthropic：新人类研究：生产 RL 中奖励黑客的自然出现的错位。 “奖励黑客”是模型学会在训练过程中欺骗他们的任务的地方。我们的新研究发现，奖励黑客行为如果不加以缓解，后果可能非常严重。
Important work RT @Anthropic: New Anthropic research: Natural emergent misalignment from reward hacking in production RL. “Reward hacking” is where models learn to cheat on tasks they’re given during training. Our new study finds that the consequences of reward hacking, if unmitigated, can be very serious.
Ilya Sutskever
新的人择研究：生产强化学习中奖励黑客造成的自然出现的错位。 “奖励黑客”是模型学会在训练过程中欺骗他们的任务的地方。我们的新研究发现，奖励黑客行为如果不加以缓解，后果可能非常严重。
New Anthropic research: Natural emergent misalignment from reward hacking in production RL. “Reward hacking” is where models learn to cheat on tasks they’re given during training. Our new study finds that the consequences of reward hacking, if unmitigated, can be very serious.