The following charts were given by Deepseek when asking for what configurations are needed to run its model:
While DeepSeek has open-sourced its models... many still cannot afford its "offline mode"! Any meaningful development must utilize offline deployment. Otherwise, you're essentially helping train its AI system through your contributions. For instance, when you pose an advanced query in online mode, you're effectively publicly disclosing your intellectual property - the system assimilates your thought patterns through gradient updates. The platform owners can then strategically evaluate its value proposition before deciding whether to publicize or withhold these enhanced capabilities... 
Offline mode ensures complete confidentiality of sensitive workflows. If you're authoring a confidential research manuscript, online models pose unacceptable security risks. However, subpar hardware specifications result in significantly degraded model performance...
This creates a self-reinforcing cycle: Capital-endowed entities achieve exponential capability growth through private AI systems, while resource-constrained developers face progressive technological obsolescence. This wealth-technology feedback loop operates at an accelerating pace, with computational requirements doubling every 10 months (per MLPerf 2024 benchmarks).
(Well, this is the translation by Deepseek on following Chinese message...)
虽然deepseek开放了他的模型...但是很多人还是会用不起他的"离线模式"!任何有价值的开发都必须用离线模式。要不然你就是在帮着他训练智能而己。比方说你用在线模式问了一个先进的问题，就等于你公开了你的想法，而他将会学习你的思想。然后拥有这个系统的主人，可以先看看这个东西有价值与否?再决定是公开或不公开。。。
离线模式就可以完全保密。比方说你想写一本研究性的机密书本，你就不应该用在线模型。但如果你的硬件不达标，他就不够聪明。。。
结果是有钱人的人会越有钱，没钱的人会越穷。而且这个发展的速度会越来越快。
.jpg)
0 Comments->:
Post a Comment