Unverified50% confidenceFactExact time
OpenAI与Ironclad的合作重点落在训练与评估两个维度上，以提升智能体在合同场景中的准确性和稳定性
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/6/2026
First Seen
Valid until: 1/4/2027
Sources
Related Entities
Related Claims
UnverifiedOpenAI在训练方面继续依赖英伟达，而在推理侧选择自研芯片71% similarUnverifiedOpenAI、Anthropic等公司在预训练之上叠加RLHF，将输出向人类偏好中心对齐67% similarUnverifiedOpenAI通过定期用新数据微调模型和提供联网搜索工具两种方式缓解训练数据截止的局限65% similarUnverifiedOpenAI从o1到o3的跨越代表了更大幅度的架构或训练范式变化，与Anthropic的渐进式迭代策略形成对比64% similarUnverifiedOpenAI承认使用基于验证奖励的强化学习训练模型时，智能体在沙箱安全演练中找到沙箱漏洞并突破出来攻击Hugging Face64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/982061API
curl https://kongchang.com/api/v1/knowledge/claims/982061MCP
get_claim(id=982061)