- Superhuman AI
- Posts
- DeepSeek speeds up R2 timeline
DeepSeek speeds up R2 timeline
ALSO: How to do competitor analysis with Deep Research

Read time: under 4 minutes
Welcome back, Superhuman. Was DeepSeek’s R1 just a weird blip in the annals of AI history — or is the Chinese startup here to stay? That all depends on what it does next. Reports suggest it’s racing to put out R2 ahead of schedule in order to keep up with US rivals.
Today’s Insights
Alibaba’s SOTA video model, Photoshop on-the-go, and Meta’s new plans
DeepSeek’s new strategy for R2
Tutorial: How to do competitor analysis with Deep Research
5 new AI tools to boost your productivity
News, memes, what’s trending on socials, and more
TODAY IN AI

Alibaba's new video model comes in three different configurations. Source: Alibaba
1. Alibaba’s new open-source video model takes VLM top slot: The Chinese e-commerce giant’s Wan2.1 just swept the competition on the industry-leading VBench, which measures things like motion smoothness, text rendering, and physics accuracy.
It beats rivals like OpenAI’s Sora, Hailuo’s MiniMax, and Luma’s Dream Machine.
Plus, its code and weights are publicly available, letting programmers create their own spinoffs — something that’s rare for models of this caliber.
Here’s a thread with some examples of what it can do.
2. Adobe drops first fully-featured Photoshop mobile app: Photographers and designers now have the full power of Photoshop right at their fingertips, literally. The image-editing app — including its suite of AI features — has landed on the iPhone, with an Android version coming soon. Users can now use Firefly to generate backgrounds, remove unwanted items, or expand a scene beyond its original frame directly in the app.
3. Zuckerberg eyes a big expansion: According to The Information, Meta is reportedly planning to build a new AI data center that could cost over $200 billion in Louisiana, Wyoming, or Texas – dwarfing the initial plans for a $10 billion center. This could be part of the company's ambitious expansion plans, which include up to $65 billion in AI infrastructure spending this year.
PRESENTED BY SANA
Build custom AI agents in minutes (not months)

With Sana, you can deploy powerful, no-code AI agents that solve end-to-end workflows and empower every team–finance, sales, support, and more.
One AI platform—unifies intelligence across your org.
Open and extensible—integrates with all AI models, enterprise apps, and cloud providers.
Strategic AI partnership—with expert solution engineers and strategists.
“We see real operational efficiency, and Sana has already provided us with a quick payback.” –Paul Laurin, CEO Piab
FROM THE FRONTIER
DeepSeek’s R2 is coming sooner than expected

DeepSeek’s is speeding up its R2 release window. Source: Getty Images
There’s been a change of plans. DeepSeek originally aimed to release its R1 successor in May, but Claude 3.7, Grok 3, and the soon-to-be-unveiled GPT-4.5 have sent the company scrambling to shorten that timeline.
In fact, China’s entire AI industry is reportedly kicking things into high gear, with Tencent, Alibaba, and ByteDance racing to get their hands on as many Nvidia H20 chips (the less-powerful ones that aren’t subject to US sanctions) as possible.
There’s a lot on the line: R1 sent a trillion-dollar shockwave through the global economy last month as investors started to question their AI spending habits. They took a big sigh of relief when OpenAI and other rivals quickly countered with impressive reasoning models of their own. But news of R2’s impending release has them holding their breaths again.
What’s at stake? If DeepSeek can drop another state-of-the-art reasoning model with just a tiny fraction of its competitors’ funding, it could send the industry into a tailspin all over again. On the other hand, if the likes of OpenAI and Anthropic can hold onto their leads, they’ll prove that scaling up with bigger and bigger data centers really was worth it all along.
THE AI ACADEMY
How to do competitor analysis using Deep Research

Go to ChatGPT and sign up with your account.
Select ‘o3-mini’ model and turn on ‘Deep Research’ feature.
Enter your prompt and press enter.
Sample Prompt: Do an analysis of the [enter name] industry. Create a list of the top 5 [enter industry name]. Analyze every aspect you can find from team size and roles, brand strength, prices, and revenue. Also list strengths and weaknesses of each [enter business name]. Be as detailed as possible. This report will be used by these [enter industry] to guide their business strategy and millions of dollars are at stake. Scope: All kinds of [enter your business niche]. Rank them based on list size and estimated revenue. Metrics: Dive into other metrics like audience demographics, engagement rates, monetization strategies, partnerships, and so on. Geographical Focus: Global but especially the [targeted market] Competitor Benchmarking: Analyze them separately and at the end create a comparison chart.
You’ll get a detailed analysis (~10k+ words) with all the key details in a few minutes.
PRESENTED BY SECTION SCHOOL
Complimentary tool: AI Pitch Deck Template

Make the case for AI to your stakeholders in this plug-and-play slide deck filled with data-backed proof points and expert-vetted frameworks for mapping AI initiatives to team ROI.
Download here.
AI & TECH NEWS
Everything else you need to know today

Anyone can now use Alphabet’s Gemini Code Assist at no cost. Source: Google
🧑💻 Copilot Crusher: Google is giving away its AI coding platform, Gemini Code Assist, at no cost, with usage limits 90 times higher than GitHub Copilot’s equivalent.
📚 Mind Multiplier: OpenAI says ChatGPT’s new deep research feature is rolling out to all Plus, Team, Edu, and Enterprise users. They’ll get 10 queries per month.
🤔 Ready and Waiting: Replit unveiled Agent v2, a new update that lets the Claude 3.7-powered coding assistant think deeply about programming decisions, boosting its overall performance.
🏭 Against the Grain: A new report suggests Microsoft could be scaling back some of its AI infrastructure projects — even as rivals like OpenAI and Apple go all-in on AI spending.
💰 Funding: Tel Aviv-based startup Quantum Machines raised $170M to help fellow tech companies get ready for the transition to quantum computing. And on the heels of its new frontier model Claude 3.7 Sonnet, Anthropic announced it’s upping its next funding round to $3.5B, giving it a $61.5B valuation. Plus, Perplexity just launched a $50 million VC fund focused on early-stage AI startups.
PRODUCTIVITY
5 AI Tools to Supercharge Your Productivity
✅ Phind: An AI search engine that uses multi-step reasoning to find the best answer for your question and a visual UI to present it in the most beautiful and interactive way.
✅ Scrybe: The AI agent that grows your LinkedIn following by helping you find viral content ideas and write posts that perform.
✅ Datadog*: Track, assess, and optimize your organization’s usage of OpenAI. Get an in-depth overview of cost estimation, prompt and completion sampling, performance metrics, and more.
✅ Zapier Agents: Create your own AI agents in minutes, equip them with company knowledge, and have them do work across 7,000+ apps.
✅ Lemonfox AI: Turn any text into high-quality, natural-sounding speech in seconds, with support for 50+ voices in eight languages.
* indicates a promoted tool, if any
PROMPT OF THE DAY
Data Visualization
Prompt: [upload data] Create a line graph to show the trend in [enter data set of interest, i.e website traffic] over the past year.
What type of prompt would you like to see next? |
Source: analyticshacker
SOCIAL SIGNALS
What’s trending on socials today

✍️ Writer’s Reckoning: Despite the AI boom sending him into an “existential crisis,” writing instructor David Perell shares his advice for how to incorporate LLMs into your writing workflow.
👾 Catch ‘Em All: One way Anthropic tested Claude 3.7’s reasoning capabilities: Watching how well it plays Pokémon. At first, it just ran away from the game’s bosses. Eventually though, it made some surprising moves, like asking for a reset after getting stuck in the corner of one level. (Here’s the full story.)
🍓 Sweet Surprise: Claude 3.7’s “Strawberry” easter egg is getting more and more elaborate.
🎚️ Level Up: One X user found a surprisingly simple prompt to make the base version of ChatGPT act more like ChatGPT Plus — although the jury’s still out on whether the hack actually works.
🧠 Knowledge is Power: Data scientist Avi Chawla breaks down (with charts) the difference between transformers and mixture of experts — two major strains of AI architecture.
AI-GENERATED IMAGES
City Walks

Source: @u3853643988 on Midjourney
Midjourney Prompt: walks around the meat packing district of NYC neighbourhood in the height of summer disposable camera aesthetic --ar 91:128 --sref 2395504882 --style raw --profile 6eykwgw
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!
What did you think of today's email?Your feedback helps me create better emails for you! |
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