Anthropic 限制部分模型研究权限，独立研究者转向开源模型
As an independent researcher, this sucks, honestly. But I guess that's where open models come in, an...
Anthropic 收紧了部分研究用例的模型访问权限，但明确保留了可解释性等领域。做 AI 安全或机制研究的独立研究者应该关心这件事，被限制的用例只能靠改权重的开源模型来做。Llama、Qwen 这类开源权重模型在自主研究场景里反而更稳。
Anthropic 近日对涉及特定敏感领域的研究用例收紧了闭源模型的访问限制，引发独立研究者的不满。研究者 Elie Bakouch 表示，自己很喜欢用 LLM 做相关主题的实验，这一限制令他感到失落，但同时肯定了 Anthropic 此次主动公开说明限制范围的做法。
与部分厂商静默收紧权限不同，Anthropic 这次明确了限制边界，可解释性（interpretability）等研究领域并未被纳入限制，相关研究仍可正常进行。受到影响的独立研究者则表示将转向开源模型，因为这类研究需要对模型权重进行修改，开源模型在透明度和可控性上反而更适合。
这一讨论凸显了闭源 API 与开源权重在研究场景中的核心差异：依赖 API 的研究者随时可能因厂商政策变化而失去工作条件，而拥有开源模型权重的研究者可以自行修改、审计和复现。对独立研究者和小型学术团队而言，权重可控的开源模型正在成为降低研究中断风险的实际选择。
As an independent researcher, this sucks, honestly. But I guess that's where open models come in, an...
As an independent researcher, this sucks, honestly. But I guess that's where open models come in, and they are actually better suited to solve that problem. For that type of work, I think it helps to own the model (change weights, etc.) and have more transparency. Again, not mad at Anthropic, and it helps that they are transparent about it this time around. x.com/eliebakouch/st… elie @eliebakouch while this personally makes me really sad as i love playing with llms on this kind of topic, i appreciate the transparency from anthropic here and also not nerfing domains like interpretability 🔗 View Quoted Tweet📊 2 ⚡
This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI regulation, compliance frameworks, and sovereign compute policy and AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Anthropic, 开源模型, AI安全, 可解释性.
Source attribution: curated from elvis (@omarsar0).
