Daily AI Newsletter: OpenAI says it has evidence China’s DeepSeek used its model to train competitor

Meta is reportedly scrambling ‘war rooms’ of engineers to figure out DeepSeek

In today’s email:

👨🏻‍⚖️ OpenAI launches ChatGPT plan for US government agencies
👀 Alibaba releases AI model it says surpasses DeepSeek
🚔 Cleveland police used AI to justify a search warrant. It has derailed a murder case
🧰 13 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

OpenAI says it has evidence China’s DeepSeek used its model to train competitor

OpenAI has accused the Chinese AI startup DeepSeek of using its proprietary models to train a competing open-source model, potentially violating intellectual property rights. The company claims to have evidence that DeepSeek engaged in "distillation," a technique where smaller models learn from larger ones by leveraging their outputs. While distillation is common in AI development, OpenAI argues that using its model’s outputs to build a rival product breaches its terms of service. Microsoft and OpenAI had previously blocked DeepSeek’s accounts suspected of engaging in such activities, but details of the investigation remain undisclosed. DeepSeek has not responded to these allegations, which were made public during the Lunar New Year holiday.

DeepSeek’s recent AI advancements have startled the industry, particularly its ability to develop powerful models at a fraction of the cost of major players like OpenAI and Google. Using just 2,048 Nvidia H800 GPUs and $5.6 million, DeepSeek trained a model with 671 billion parameters, raising suspicions that it leveraged OpenAI’s outputs to achieve comparable results. AI experts note that many startups and academic researchers rely on responses from advanced AI systems like ChatGPT to train their own models, effectively bypassing the expensive human feedback process. This practice, while widespread, presents challenges for companies trying to protect their technological edge.

OpenAI emphasized that it actively deploys countermeasures to safeguard its intellectual property, warning that Chinese and other foreign firms consistently attempt to extract knowledge from leading US AI models. The company also highlighted its collaboration with the US government to prevent adversaries from exploiting American AI technology. The controversy underscores broader concerns about intellectual property in AI, with OpenAI itself facing lawsuits from content creators, including The New York Times and several authors, over claims that it used copyrighted materials without permission to train its own models.

Free Notion and Unlimited AI

Thousands of startups use Notion as a connected workspace to create and share docs, take notes, manage projects, and organize knowledge—all in one place. We’re partnering with Notion to offer 3 months of new Plus plans($3000 value), including unlimited Notion AI so you can try it all for free!

To redeem the Notion for Startups offer:

Submit an application using our custom link: https://ntn.so/superpowerdaily and select Superpower Daily on the partner list.
Include our partner key, STARTUP4110P37060.

Free Notion and Unlimited AI

Meta is reportedly scrambling ‘war rooms’ of engineers to figure out how DeepSeek’s AI is beating everyone else at a fraction of the price

Meta has reportedly assembled four “war rooms” of engineers to investigate how Chinese AI startup DeepSeek, backed by High-Flyer Capital Management, has managed to develop an AI chatbot, R1, that rivals top models like ChatGPT at a significantly lower cost. This open-source breakthrough has raised concerns about the massive AI investments made by U.S. companies and has put Meta on high alert. According to The Information, Meta AI infrastructure director Mathew Oldham has warned colleagues that DeepSeek’s newest model may even outperform the upcoming version of Meta’s Llama AI, which is expected to launch in early 2025.

Two of Meta’s war room teams are focused on uncovering how DeepSeek reduced the cost of training and running its model, aiming to apply similar strategies to improve Llama’s efficiency. The other two teams are investigating DeepSeek’s training data sources and exploring ways to restructure Meta’s models based on DeepSeek’s architecture. While Meta has not officially commented, a company spokesperson emphasized that Meta continuously evaluates competitive models and remains committed to leading the open-source AI ecosystem with the upcoming release of Llama 4.

Meanwhile, Meta CEO Mark Zuckerberg has announced plans to invest up to $65 billion in AI-related projects this year, including a massive data center expansion and additional AI hires. This comes shortly after OpenAI, in collaboration with SoftBank, Oracle, and others, revealed a $500 billion AI infrastructure project, Stargate, which has White House backing and aims to construct dozens of new data centers across the U.S. As competition in the AI space intensifies, Meta is racing to ensure that Llama remains competitive against emerging challengers like DeepSeek.

OpenAI launches ChatGPT plan for US government agencies

ChatGPT Gov is a specialized version of ChatGPT designed to provide U.S. government agencies with secure and efficient access to OpenAI’s cutting-edge AI models. By leveraging Microsoft Azure’s commercial and government cloud environments, agencies can deploy ChatGPT Gov while maintaining compliance with stringent cybersecurity standards such as IL5, CJIS, ITAR, and FedRAMP High. This self-hosted infrastructure ensures agencies can securely handle sensitive data while benefiting from AI-driven tools for improving public services, national security, and administrative efficiency. The platform includes key features from ChatGPT Enterprise, such as conversation sharing, file uploads, GPT-4o access, Custom GPTs, and an administrative console for IT management.

Since 2024, over 90,000 users across 3,500 U.S. federal, state, and local agencies have integrated ChatGPT into their workflows, enhancing efficiency in areas like scientific research, translation services, and administrative automation. The Air Force Research Laboratory uses ChatGPT Enterprise for coding support and AI education, while Los Alamos National Laboratory employs it for bioscientific research. The State of Minnesota’s Enterprise Translations Office has significantly improved multilingual service delivery, and a pioneering AI pilot program in Pennsylvania has demonstrated notable time savings for government employees, cutting down routine tasks by approximately 105 minutes per day.

OpenAI remains committed to supporting the U.S. government in harnessing AI’s potential to improve public sector operations while aligning with democratic values. As efforts continue toward achieving FedRAMP Moderate and High accreditations for ChatGPT Enterprise, OpenAI is also exploring expansion into classified cloud regions. By working closely with government agencies, OpenAI aims to foster responsible AI adoption, enhance service delivery, and build public trust in artificial intelligence as a transformative tool for governance.