How To Use DeepSeek On A US Server

In partnership with

Good morning. It’s Wednesday, January 29th.

Did you know: On this day in 1999, Yahoo bought GeoCities for $3.65 billion, only to shut it down 10 years later.

In today’s email:

  • DeepSeek now offered on Perplexity with US/EU Servers

  • Did DeepSeek Steal OpenAI Data?

  • Chinese Robot Maker Eyes Mass Production

  • 3 New AI Tools

  • Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

The future of presentations, powered by AI

Gamma is a modern alternative to slides, powered by AI. Create beautiful and engaging presentations in minutes. Try it free today.

Today’s trending AI news stories

Perplexity Now Offers DeepSeek AI Based in US/EU Servers

DeepSeek R1 is now live on Perplexity AI. Perplexity’s new Pro Search mode lets users switch between DeepSeek R1 and OpenAI’s o1, complete with chain-of-thought transparency. Daily usage limits have been upped for both free and paid tiers, all while keeping data firmly anchored in US/EU servers.

Perplexity CEO Aravind Srinivas hailed the integration, calling it a leap forward for AI search agents and vowing to scale US data center capacity to meet demand. As Srinivas put it, “The proliferation of search agents and assistants that can reason has just begun.” Read more. (Paywall)

A group connected to DeepSeek reportedly exfiltrated a large amount of data using OpenAI’s application programming interface

Microsoft and OpenAI are investigating whether DeepSeek, a Chinese AI startup, improperly leveraged OpenAI's API to siphon proprietary data at scale. Microsoft’s security team reportedly observed individuals tied to DeepSeek extracting large volumes of data via OpenAI’s API last fall, potentially violating OpenAI’s terms of service.

The allegations, including claims of “distillation” techniques using OpenAI’s outputs, have stirred market volatility, erasing nearly $1 trillion from tech stocks. Achieving tenfold efficiency over competitors like Meta, this milestone was made possible by bypassing industry-standard Nvidia CUDA framework in favor of its lower-level, assembly-like PTX programming, according to an analysis from Mirae Asset Securities Korea, cited by @Jukanlosreve. This approach enabled granular optimizations, such as reconfiguring streaming multiprocessors and refining thread-level operations.

However, concerns arise over whether DeepSeek’s approach signals a shift towards more software-driven efficiency or remains an isolated feat. Industry figures remain divided: some view the achievement as a step towards broader AI accessibility, while others argue the hardware requirements for advanced models will persist regardless of optimization progress. (Paywall bypass link) Read more.

Chinese robot maker UBTech eyes mass production of industrial humanoids by year end

UBTech Robotics is gearing up to scale industrial automation, with plans to mass-produce its Walker S Series humanoid robots by late 2025. The Shenzhen-based firm aims to deploy 500 to 1,000 units this year, targeting heavyweights like Foxconn, automotive manufacturers, and logistics leader SF Express.

Factories offer structured environments ideal for training these bots, while labor shortages make the case for automation undeniable. Though the dream of household humanoids remains on the horizon, current AI limitations keep the focus on industrial use cases. Since its 2012 launch, UBTech has expanded into non-humanoid bots for cleaning, delivery, and service roles. The Walker S1, unveiled in late 2024, underscores its push to dominate China’s cutthroat robotics arena. Read more.

5 new AI-powered tools from around the web

arXiv is a free online library where researchers share pre-publication papers.

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!