Gemini 2.0 Pro is here

Read time: under 4 minutes

Welcome back, Superhuman. Google was never going to sit and watch from the sidelines as OpenAI dropped model after model. The search giant just dropped its latest family of Gemini 2 models — and they pack a serious punch.

Today’s Insights

• Today in AI: Cursor’s hockey stick growth

• Frontier: Google unveils ‘world’s best’ LLM

• Tutorial: How to do research analysis with Deep Research

• 5 new AI tools to boost your productivity

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

TODAY IN AI

1. Wrappers may have the last laugh: AI-powered code editor Cursor is now reportedly the fastest app in history to go from $1M to $100M in revenue. The news comes at an interesting time, as foundation models appear to have little defensibility in the face of a new wave of cheaper and open-source models. But apps that were written off as “GPT wrappers“ a few months ago may be more durable than many expected.

2. Meta masters motion with new AI video technique: VLMs still struggle with action shots — sometimes inadvertently turning a gymnast’s routine or a talent show into surreal clips from a horror movie. Meta’s solution: Training a model from the ground up to focus on motion. That approach can be applied to any VLM “with minimal adaptations,” to achieve state-of-the-art results. (See how it compares to rivals like Sora, Kling 1.5, and Runway’s Gen3 here.)

3. Google’s new flagship model sweeps the competition: The search giant just unveiled Gemini 2.0 Pro Experimental, which took the top spot across all categories in the industry-leading LMArena. Meanwhile, a new reasoning model rivaling DeepSeek’s R1 and OpenAI’s o3-mini is now available to all users in the Gemini app. Keep reading for all the details.

A WORD FROM OUR SPONSORS

Next 2025 Startup Hub

Your go-to place for immersive showcases, one-on-one Google Cloud expert guidance, and networking opportunities with fellow entrepreneurs and global VCs.

Learn, connect, and be inspired – all while having fun. Get your Google Cloud Next 2025 ticket at a lower price the earlier you register, and join us April 9–11 at Mandalay Bay Convention Center in Las Vegas.

For startups, we have an exclusive discount code to get your tickets at $799/ea, just use code “799STARTUP”

FROM THE FRONTIER

Anyone can now try Google’s ‘most capable’ model

Google just put one of the "world’s best AI models” (as judged by the LMArena) into everyone’s hands. Gemini 2.0 Flash Thinking Experimental can break complicated tasks down and explain its thought process along the way, similar to OpenAI and DeepSeek’s latest reasoning LLMs. And anyone with a Google account can try it out on the Gemini app.

That’s just one of several models unveiled Wednesday:

• 2.0 Flash Thinking Experimental with apps: This variant can connect with apps like YouTube, Maps, and Search, unlocking new capabilities.

• 2.0 Pro Experimental: This one’s best for coding, math, and other complex prompts, with a massive 2M-token context window. It’s only available to developers and Gemini Advanced users for now though.

• 2.0 Flash-Lite: Another model geared toward developers. It performs better than its predecessor but with the same cost and speed — and it’s multimodal, which lets you do things like generate captions for thousands of photos at a time.

• Google plans to spend about $75B this year, much of it on AI, proving that DeepSeek’s R1 hasn’t totally upended the market as some had initially expected.

Why it matters: Just a few months ago, top reasoning models were only available to hardcore users willing to pay for subscriptions and compute. Google’s latest releases mark another step toward truly accessible AI, letting anyone conduct in-depth research, build their own agents, and deploy coding projects at no cost. With progress speeding up by the day, it’s only a matter of time before OpenAI or Anthropic respond with something even better.

THE AI ACADEMY

How to do research analysis using Deep Research

• Go to ChatGPT and sign up with your account. *Note: Pro users have access to 100 deep research tasks per month. Plus, Team, Enterprise, and Free users will get access at a later date, according to OpenAI.

• Now select ‘o3-mini’ as your model at the top.

• Turn on the ‘Deep research’ button near the prompt box.

• Give it a prompt, and it will autonomously search the web (or provided sources) to compile an answer.

Sample Prompt: ‘Conduct a comprehensive analysis of the impact of AI-generated content on traditional media industries. Provide recent examples from 2023-2024 where AI-generated news, images, or videos have caused significant controversy or shifts in media consumption. Analyze industry reports, case studies, and expert opinions to highlight key trends. Compare how different countries are responding with policies or regulations. Include insights from media executives, journalists, and AI ethicists on the future of journalism in an AI-dominated landscape.’

• It will take 5 to 30 minutes to return long and well-researched reports from many different sources.

PRESENTED BY HYPHENATE

Need to win more proposals, faster? Hyphenate can help

No matter your industry, Hyphenate can help you create unbeatable proposals and SOWs–30x faster than by hand.

• Understands the needs of your business and client

• Quality guardrails ensure content is highly customized and delivered in your voice

• Initial draft ready in minutes

Schedule a demo today.

AI & TECH NEWS

Everything else you need to know today

🖼️ Visual Learner: DeepSeek just added VL2-Small, a model that’s especially good at analyzing documents and images, to Hugging Face as a demo for developers to try out.

🔍 Search Sidekick: ChatGPT search, which lets the chatbot access real-time information from the web, is now available to everyone — no sign-up needed.

🗣️ Long Time Coming: Amazon is reportedly getting ready to release an AI-powered Alexa in a few weeks. It was built using a unique architecture that makes hallucinations far less likely.

💭 Dream Team: Microsoft’s Mustafa Suleyman scooped up three of his former DeepMind colleagues, who’ll bring their expertise building NotebookLM’s podcast tool to Microsoft’s upcoming Astra project.

🛒 AI Emporium: Hugging Face’s Spaces platform got a major overhaul, letting developers easily browse 400K apps by category — like an app store built specifically for AI releases.

📈 Markets: Nvidia closed Wednesday up 5.21% after Alphabet and other big tech companies doubled down on their plans to spend big on AI this year. But Alphabet’s own stocks fell 6.94% over concerns that the tech giant might not be able to translate that spending into revenue.

PRODUCTIVITY

5 AI Tools to Supercharge Your Productivity

✅ a0.dev: Build mobile apps in days instead of weeks by generating full react native apps and custom components with AI.

✅ Casafy AI: An AI search engine that analyzes properties based on their physical characteristics, allowing businesses to define and search for any visual criteria they need.

✅ Concierge*: AI Assistant that talks to your apps in real time. Ask it to update HubSpot, create Jira tickets, summarize Slack, search Confluence, reply to emails, and more.

✅ Jace: An AI email agent that researches emails, calendars, web, and files to craft responses and schedule meetings.

✅ Skyvern: An open-source agent capable of browsing the web and taking action, coupled with a no-code tool to help you build your own agents.

^{* indicates a promoted tool, if any}

PROMPT OF THE DAY

LinkedIn Experts

Prompt: Generate a LinkedIn post where I share my insights on [insert topic here]. Keep the content engaging and easy to understand. The post should consist of 200 words maximum. Also include relatable SEO hashtags.

_{Source: Careerflow}

SOCIAL SIGNALS

What’s trending on socials today

👵 Scammer’s Nightmare: Meet “Daisy,” a model meant to resemble an elderly grandmother who doesn’t know how to work her phone. She’s being used in the UK to trap phone scammers in endless conversation loops.

📚 Speedy Scholar: A George Mason professor says OpenAI’s deep research tool is like “having a good PhD-level research assistant, and sending that person away with a task for a week or two” — except it can generate results in as little as five minutes.

🤖 Cyber Colleague: Economist Joshua Gans claims he collaborated with GPT-o1-pro to transform one of his ideas into a fully formed academic paper in only an hour — then got it published in a peer-reviewed journal.

🎒 Study Up: OpenAI cofounder Andrej Karpathy just dropped a three-and-a-half hour YouTube tutorial about how LLMs are trained — “along with mental models of how to think about their ‘psychology.’”

AI-GENERATED IMAGES

Group photo

Midjourney Prompt: collage of a dog park. Include only golden brown colored dogs. Navy blue and gold with a small slash of deep purple color palette. bold colors, uplifting mood. --chaos 3 --ar 9:16 --stylize 1000 --profile mh3xhwo

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