DeepSeek's R1: Innovation or imitation?

Read time: under 4 minutes

Microsoft claims that DeepSeek might have helped itself to OpenAI's data without permission. But in an industry where everyone learns from everyone else, drawing the line between inspiration and imitation isn't so simple.

Also: Supersonic travel looks set to make a comeback, and learn how to use DeepSeek’s R1 securely in the US.

Today’s Insights

• Today in AI: Coding agents, deep sea discoveries, and Meta’s conundrum

• Tutorial: How to use DeepSeek R1 securely in the US

• Frontier: Another DeepSeek mystery

• 5 new AI tools to boost your productivity

• News, memes, what’s trending on socials, and more

TODAY IN AI

1. Twitter co-founder returns with open-source AI agent: Jack Dorsey’s Block, the company behind Cash App and Square, is wading into the world of agents with a coding-focused tool called Goose. While most agents — say, OpenAI’s Operator or Alphabet’s Project Astra — are built on top of specific models, Goose works across many different apps. Instead of having a lengthy back-and-forth with your chatbot, you can give Goose a simple prompt and let it do all the troubleshooting for you.

2. Stuck in the middle with Meta: Meta took a big risk early on by opening up its models for all to see. Can it stay true to its vision now that there’s a new sheriff in town? After all, R1 gives the company a new blueprint to work off of, and it’s already set up “war rooms” to investigate the model’s inner workings. CEO Mark Zuckerberg even said DeepSeek’s rise “has only strengthened our conviction that this is the right thing for us to be focused on.” Proponents say that in the end, it’s not Meta but its closed-source rivals who should be breaking a sweat.

3. AI-powered robots are spilling the ocean’s secrets: We know more about Mars than we do about the deep sea — but that may be about to change. Scientists are testing an AI-powered underwater robot, MiniROV, that can autonomously track rarely-seen jellyfish, beaming the images to the surface for further study. The model is being trained by FathomVerse, a mobile game where players spot deep-sea organisms. This citizen science effort has already logged nearly 50,000 IDs.

PRESENTED BY DEEPGRAM

Voice AI Platform for Developers

Build production-ready voice applications in minutes, not months with Deepgram.

• Powerful APIs for speech-to-text, text-to-speech, and AI voice agents

• Reliable, low-latency, affordable

• Built for developers, ready for enterprise

Join 200K developers building with Deepgram. Sign up

THE AI ACADEMY

How to use DeepSeek R1 securely in the US

• Go to Perplexity and log in with your account.

• Now click on the ‘Pro’ button near the prompt box and a dropdown will appear.

• Select ‘Reasoning with R1’ as your model.

• Enter your prompt and press enter to get results.

• Perplexity will proceed to generate the output from the R1 model.

(NOTE: If you want to run more than 5 searches per day, you’ll have to purchase a Perplexity Pro subscription)

FROM THE FRONTIER

Did DeepSeek cheat? It’s complicated.

The latest twist in the DeepSeek saga: Microsoft says it found evidence that the Chinese startup swiped OpenAI’s proprietary data without permission (although it’s already added R1 to its own cloud offerings). Users can pay for access to some of OpenAI’s data, but the claim is that a mysterious group with ties to DeepSeek grabbed way more than was allowed sometime last fall.

First, some context: Models like GPT-4 and Claude 3.5 are trained on trillions of data points. But as AI companies run out of data to feed their LLMs, they’re turning to a tactic called reinforcement learning. It involves rewarding models when they make the right decisions, and it’s led to a sudden jump in capabilities.

What’s the big deal? The problem is that DeepSeek allegedly got to skip past all the hard parts by putting the finishing touches on an architecture that OpenAI spent at least millions of dollars developing. It might have used a method called distillation, when you feed outputs from a large model into a much smaller one to train it at a much faster pace.

The news has divided the industry: Some critics think OpenAI had it coming after using copyrighted books and webpages for its own models. Others, including US AI czar David Sacks, say this is no time for gloating. They argue the alleged breach brings up major security concerns and could open the floodgates for more copycats — undermining the US’s lead in the process.

PRESENTED BY BAMBOOHR

65% of New Hires Have No Clear Points of Contact for Questions

23% of new hires cry in their first week—often from imposter syndrome, info overload, or negative interactions. Onboarding should ease anxieties, not add to them!

Discover how to boost retention, meet employee expectations, and create an inclusive experience. Download the Definitive Guide to Onboarding today

AI & TECH NEWS

Everything else you need to know today

📈 Bot Boom: AI-powered mobile apps are now a $2B market, with ChatGPT leading the pack. The app has grossed $529M since it launched in May 2023 — nearly 4x the second-highest earner.

💵 Cash Cow: ChatGPT Pro’s $200/month subscription is pulling in $300M a year, giving OpenAI a much-needed revenue boost. Meanwhile, a new study shows the app is dominated by males, who make up nearly 85% of all users.

🧑‍🎨 Human Touch: In a major new report, the US Copyright Office clarified that artists are allowed to copyright materials that were created with the help of AI — as long as human creators had a hand in the final product.

📈 Markets: Tesla closed Wednesday down 2.26% after a disappointing earnings report showed that it had only brought in 2% higher revenue than the year before. Microsoft suffered a similar fate, finishing down 1.09% due to lower-than-expected cloud revenue. Meanwhile, Nvidia shares slid 4.1% after the White House announced it’s considering stricter export controls on the chip giant’s GPUs.

💰 Funding: UVeye, which uses AI to help speed up vehicle inspections, raised $191M in an extension of its Series D funding round. The startup is building a vision-based model that can perform “MRIs for cars” for clients including Amazon and CarMax.

PRODUCTIVITY

5 AI Tools to Supercharge Your Productivity

✅ Jasper: A purpose-built AI that helps enterprise marketers build AI-powered apps and workflows tied to real business outcomes.

✅ Baz: An AI-powered coding platform leveraging specialized models and embeddings to address the limitations of traditional coding tools.

✅ Eightsleep*: The Eight Sleep Pod uses sophisticated algorithms to regulate body temperature and reduce snoring throughout the night. Can be added to any mattress.

✅ Overseer AI: Validate AI content instantly with a single API call, helping you ship safer products, faster.

✅ Wepost: Simplify your social media workflow with an AI tool that automates content creation, publishing, and analytics, so you can focus on building your brand.

^{* indicates a promoted tool, if any}

PROMPT OF THE DAY

Product Roadmap

Prompt: You are a product manager working on [insert product or project]. Create an effective product roadmap and prioritize features. Discuss techniques such as user story mapping, stakeholder collaboration, and data-driven prioritization to ensure the roadmap aligns with customer needs and business goals.

_{Source: Sid Saladi / Substack}

SOCIAL SIGNALS

What’s trending on socials today

🛩️ Flying High: Supersonic travel looks set to make a comeback after Boom Supersonic’s successful flight. Their CEO explained why the Concorde failed, and why Boom won’t.

🤖 Wrap Battle: The term “GPT wrapper“ may no longer be considered a derogatory term in Silicon Valley, as foundational models face the risk of being commoditized and products on the app layer look well-positioned to capture value.

📜 Waxing Poetic: DeepSeek’s R1 went deep — maybe too deep — after an X user prompted it to “write a heart-rending piece of free poetry about what it means to be an AI in 2025?”

🔥 Hot Take: Anthropic CEO Dario Amodei weighed in on the DeepSeek debacle, arguing that R1’s capabilities, while impressive, aren’t all that surprising in light of the industry’s overall progress.

🧠 Thrifty Thinking: A research team from Berkeley claims to have recreated a small version of R1 for less than the cost of a nice dinner.

Acquire new customers and drive revenue by partnering with us

Superhuman is the world’s biggest AI newsletter for businesses and professionals with 1M+ readers and 2M+ followers on socials working at the world’s leading startups and enterprises. Companies like Amazon, Hubspot, and Salesforce feature their products in Superhuman. You can learn more about partnering with us here.

Your opinion matters!

Got more feedback or just want to get in touch? Reply to this email and we’ll get back to you.

Thanks for reading.

Until next time!

Zain & the Superhuman AI team