内容来源：https://techcrunch.com/2025/09/10/thinking-machines-lab-wants-to-make-ai-models-more-consistent/

内容总结：

由米拉·穆拉蒂领导的Thinking Machines实验室近日首次披露其重点研究方向：构建可生成确定性响应的人工智能模型。该实验室凭借20亿美元种子融资及OpenAI前研究员组成的明星团队，已成为硅谷最受瞩目的AI初创企业之一。

在题为《破解LLM推理中的非确定性难题》的研究博客中，研究员贺瑞斯（Horace He）指出，当前AI模型响应随机性的根源在于GPU内核调度方式。通过精确控制芯片层面的运算协调，有望使ChatGPT等AI工具每次对相同问题给出高度一致的答案。

这项技术突破不仅能为企业和科研机构提供更可靠的AI响应，还将显著提升强化学习训练效果——当模型输出保持稳定时，奖励机制的数据质量将得到优化。据此前报道，该实验室计划运用强化学习技术为企业定制AI模型。

作为OpenAI前首席技术官，穆拉蒂曾表示实验室首款产品将于数月内面世，重点服务研究人员和初创企业的模型开发需求。虽然尚未确认该产品是否会应用此次发布的确定性技术，但实验室承诺将持续公开研究成果，这一开放态度与逐渐封闭的OpenAI形成鲜明对比。

目前该实验室正以120亿美元估值进行融资，其能否攻克AI前沿领域的核心难题，并将研究成果转化为实际产品，将成为对其估值合理性的关键考验。

中文翻译：

米拉·穆拉蒂的"思维机器实验室"正以其20亿美元种子资金和OpenAI前研究员组成的全明星团队开展研究，外界对其研发内容始终抱有浓厚兴趣。本周三该实验室通过博客文章首次向外界展示了其重点项目之一：构建具有可重现响应能力的人工智能模型。

这篇题为《攻克LLM推理中的非确定性难题》的研究博客试图揭示AI模型响应随机性的根源。例如多次向ChatGPT提出相同问题，往往会得到截然不同的答案。AI界普遍将这一现象视为既定事实——当今的AI模型被认为属于非确定性系统——但思维机器实验室则认为该问题可以被攻克。

文章作者、实验室研究员贺拉斯·赫指出，AI模型随机性的根本原因在于GPU内核（英伟达计算机芯片内部运行的小型程序）在推理处理过程中（即在ChatGPT中按下回车键后发生的所有计算）的拼接方式。他提出通过精细控制该协调层，有望使AI模型获得更强的确定性。

赫强调，除了为企业和科研人员提供更可靠的响应外，实现AI模型的可重现响应还能改进强化学习训练效果。强化学习是通过奖励正确答案来训练AI模型的过程，但当答案存在细微差异时，训练数据就会产生噪声。他认为创造更一致的AI响应能使整个强化学习过程"更流畅"。据The Information此前报道，该实验室已向投资者表示计划运用强化学习技术为企业定制AI模型。

OpenAI前首席技术官穆拉蒂曾在七月表示，思维机器实验室的首款产品将于数月内亮相，该产品将"对开发定制模型的研究人员与初创企业具有实用价值"。目前尚不清楚具体产品形态，以及是否会采用这项研究的技术来实现更可重现的响应。

该实验室还承诺将定期发布博客文章、代码及其他研究成果，旨在"既造福公众，也提升自身研究文化"。本次作为新开设"连接主义"系列博客的首篇文章，正是这一承诺的实践。OpenAI创立初期也曾致力于开放研究，但随着规模扩张逐渐转向封闭。穆拉蒂的实验室能否践行开放承诺，仍需持续观察。

这篇研究博客为外界提供了窥视硅谷最神秘AI初创企业的罕见窗口。虽然技术发展方向仍未完全明朗，但显示出该实验室正在攻坚AI研究前沿领域的若干核心难题。真正的考验在于思维机器实验室能否解决这些难题，并基于研究成果打造出匹配其120亿美元估值的产品体系。

英文来源：

There’s been great interest in what Mira Murati’s Thinking Machines Lab is building with its $2 billion in seed funding and the all-star team of former OpenAI researchers who have joined the lab. In a blog post published on Wednesday, Murati’s research lab gave the world its first look into one of its projects: creating AI models with reproducible responses.
The research blog post, titled “Defeating Nondeterminism in LLM Inference,” tries to unpack the root cause of what introduces randomness in AI model responses. For example, ask ChatGPT the same question a few times over, and you’re likely to get a wide range of answers. This has largely been accepted in the AI community as a fact — today’s AI models are considered to be non-deterministic systems— but Thinking Machines Lab sees this as a solvable problem.
The post, authored by Thinking Machines Lab researcher Horace He, argues that the root cause of AI models’ randomness is the way GPU kernels — the small programs that run inside of Nvidia’s computer chips — are stitched together in inference processing (everything that happens after you press enter in ChatGPT). He suggests that by carefully controlling this layer of orchestration, it’s possible to make AI models more deterministic.
Beyond creating more reliable responses for enterprises and scientists, He notes that getting AI models to generate reproducible responses could also improve reinforcement learning (RL) training. RL is the process of rewarding AI models for correct answers, but if the answers are all slightly different, then the data gets a bit noisy. Creating more consistent AI model responses could make the whole RL process “smoother,” according to He. Thinking Machines Lab has told investors that it plans to use RL to customize AI models for businesses, The Information previously reported.
Murati, OpenAI’s former chief technology officer, said in July that Thinking Machines Lab’s first product will be unveiled in the coming months, and that it will be “useful for researchers and startups developing custom models.” It’s still unclear what that product is, or whether it will use techniques from this research to generate more reproducible responses.
Thinking Machines Lab has also said that it plans to frequently publish blog posts, code, and other information about its research in an effort to “benefit the public, but also improve our own research culture.” This post, the first in the company’s new blog series called “Connectionism,” seems to be part of that effort. OpenAI also made a commitment to open research when it was founded, but the company has become more closed off as it’s become larger. We’ll see if Murati’s research lab stays true to this claim.
The research blog offers a rare glimpse inside one of Silicon Valley’s most secretive AI startups. While it doesn’t exactly reveal where the technology is going, it indicates that Thinking Machines Lab is tackling some of the largest question on the frontier of AI research. The real test is whether Thinking Machines Lab can solve these problems, and make products around its research to justify its $12 billion valuation.