The many roads to AGI

Welcome back, Superhuman. It happened again. Another Chinese company has come out of nowhere with an open-source model that rivals the best from Silicon Valley. Meanwhile, OpenAI just fumbled a major deal — and Google was more than happy to snatch it up.

Also Today: Learn how to 10x productivity by creating AI apps with DataButton, and get the latest prompts, headlines, and social posts.

TODAY IN AI

1. Moonshot’s new open-source model stuns AI community: We might be witnessing another DeepSeek moment. China’s Moonshot AI just unveiled Kimi K2, featuring 1T parameters (with 32B active) and frontier performance across multiple coding and knowledge benchmarks, even outperforming Claude 4 Opus and GPT-4.1 on some tasks. That’s all the more impressive when you consider that Moonshot is relatively unknown outside of China, and K2 is completely open source, letting anyone build on top of it or probe its inner workings.

2. Windsurf CEO heads to Google, dealing major blow to OpenAI: OpenAI just missed its chance to buy Windsurf (f.k.a. Codeium) after offering to acquire the vibe coding startup for $3B a few months ago. Adding insult to injury, Google announced that it had poached the company’s CEO Varun Mohan, along with several other key employees. They’ll join DeepMind to beef up Gemini’s agentic coding capabilities.

3. Goldman Sachs ‘hires’ first AI coder: The world’s second-largest investment bank announced a new partnership with Cognition, whose Devin program helped kick off the vibe coding wave last year. JPMorgan Chase and Morgan Stanley have AI initiatives of their own. But Goldman is supposedly going a step further by letting Devin take care of entire engineering workflows autonomously, instead of just serving as an assistant for humans.

PRESENTED BY SLACK

Must-learn feature: Find every answer with Slack enterprise search

Slack enterprise search connects your business tools, conversations, and data into a single, unified search directly in Slack. 

Search across internal and external sources for every answer – without switching apps. 

• Eliminate knowledge barriers between teams 

• Surface company data instantly and securely

• Connects all your apps: Salesforce, Google Drive, Microsoft Teams, and more

Give your team real-time access to the information they need right now. 

Download the e-book to learn how.

FROM THE FRONTIER

The many roads to AGI

While meant to be a joke, the post above actually brings up a good point: While most AI startups have the same goal — being the first to reach AGI or ASI — they’re going about it in very different ways.

• OpenAI found success by scaling up compute — throwing more GPUs and training data at an LLM to make it smarter.

• xAI tweaked that approach with Grok 4 by focusing on reinforcement learning (i.e., giving a model a reward each time it gets an answer right). Rather than putting all of its compute into the initial training phase, it scaled up the RL phase instead.

• Moonshot took yet another approach when building Kimi K2. While most models have trouble “paying attention” during training, K2 used a special tool known as a Muon optimizer that helped it refresh its attention mechanism to stay sharp throughout the process.

• Meta has taken maybe the riskiest approach by offering pay packages well above the industry average. It thinks that bringing together some of the best minds in AI will lead to new techniques that haven’t even been thought up yet, although former researchers have called out the company for lacking a clear vision or direction.

The big takeaway: AI progress is no longer just about bigger models or more GPUs. It’s about scaling the right things at the right time — whether that’s compute, fine-tuning techniques, data curation, or the people behind the models. It means the next real breakthrough might not come from scaling at all, but from rethinking what we’re optimizing for in the first place.

THE AI ACADEMY

How to 10x productivity by creating your own AI apps with DataButton

Build an AI invoice processor with DataButton and eliminate HOURS of manual data entry.

• Go to DataButton, login, and ‘Create New App’

• In the description field, clearly describe your tool idea.

Sample prompt: “An AI-powered app that reads PDF invoices and automatically extracts all the data.

User uploads an invoice PDF → AI extracts vendor, amounts and line items → stores as a record and attaches the PDF → Review and export to Excel.

For small businesses tired of manually entering invoice data into spreadsheets or QuickBooks.”

• You can add requirements, specifications, quotes, user stories, and sample UI’s

• Submit and DataButton will create an app build plan. You’ll see a list of initial tasks, and can click ‘Start task’ to begin

• Chat with the agent to clarify or refine tasks throughout the build out.

• Once all tasks are completed, you’ll get a fully functional AI-powered invoice processing app

• DataButton is the ultimate full-stack AI app builder to 10x your productivity and create impactful tools like email digests, recap bots, focus schedulers, and more.

PRESENTED BY DATADOG

Free Cheatsheet: Calculate your OpenAI ROI

Datadog’s OpenAI Cheatsheet gives you crystal-clear visibility into your API costs, errors, and performance.

In just a few minutes, this guide will help you…

• Estimate OpenAI costs across libraries (Python, Node.js, PHP)

• Analyze prompts and completions

• Monitor latency/throughput and catch errors fast

Download the cheatsheet here (costs nothing), and see how cost-effective your workflow is.

IN THE KNOW

What’s trending on socials and headlines today

🗯️ Never Enough: It’s amazing how fast we become unimpressed with groundbreaking technology and start demanding more. This “Whiplash”-inspired video has racked up hundreds of millions of views by perfectly capturing that feeling.

👾 One and Done: Replit’s Matt Palmer shows how to build and deploy a simple game with Grok 4 and Replit in under two minutes. And here’s another example that’s going viral.

🏙️ From the Ground Up: Watch a wooden hut transform into a bustling metropolis in this AI-generated time-lapse, created with Hailuo AI and Suno (prompt included).

✨ Also Worth Checking Out: Grok tackles calculus, AI video virality explained, and GPT-5 rumors. 

Meta acquired PlayAI in a bid to integrate its life-like generative voice agents into Meta AI, AI Characters, and more. It’s a sign that Meta’s aggressive buying spree isn’t over yet.

Sam Altman announced OpenAI is delaying the release of its long-awaited open-source model, which was originally slated to launch this week, so it can go through more safety testing.

Grok 4 appears to study Elon Musk’s posts before responding to controversial political topics, raising doubts about xAI’s claim that the chatbot is “maximally truth-seeking.”

PRODUCTIVITY

5 New AI Tools to Boost Your Productivity

• 𖣂 Boomi Agentstudio*: The only full agent lifecycle management solution to simply design, govern, and orchestrate all AI agents across your organization. Try Boomi Today.

• 📚️ 15minutes: Get 15 minute book summaries, analysis, and more.

• 📱 AppStruct: Create apps and websites with zero coding skills.

• 🖼️ Twistly: Create slides in seconds 
with ChatGPT for PowerPoint.

• 💻 Sim Studio: Build and deploy agentic workflows with AI.

^{* indicates a promoted tool, if any}

PROMPT STATION

Create Viral Hooks

Prompt: You are a viral video strategist trained in neuroscience and copywriting.
Give me 5 high-impact video hooks (under 10 seconds) based on brain triggers like: pattern interrupt, novelty, open loops, emotional salience, or reward prediction.
Each hook must:
• Be spoken-friendly (Reels, Shorts, TikTok)
• Trigger curiosity or emotion fast
• Include the name of the brain principle used
Steps:
1. Ask for my topic + audience
2. Generate 5 hooks with labeled brain triggers
3. Help script the full 20s video if I like one
4. Offer alternates if needed
Start with:
"What's your topic and audience?"

Source: godofprompt

Sushi Room

ChatGPT Prompt: A photorealistic bedroom where every element is crafted entirely from sushi ingredients. The walls, floor, and ceiling are all made from layered sushi rolls, showcasing rice, fresh fish, and seaweed textures.

The bedframe is constructed from large, sturdy sushi rolls, supporting a mattress made from soft rice blocks layered with thin sashimi slices. The pillows and cushions on the bed are also made from carefully folded sushi pieces, resembling plush comfort.

A large wardrobe stands against one wall, made from stacked maki rolls and nigiri pieces, with sliding doors crafted from translucent seaweed sheets. The floor is covered with a sushi-patterned rug, composed of various sushi rolls arranged artistically.

Whenever you’re ready to take the next step

• Check out our Top 125 AI Tools

• Get better at prompting with our Top 1,000 Prompts

• Learn how to use AI at work and get certified with our free course

Grow customers & revenue: Join companies like Amazon, Hubspot, and Salesforce. Showcase your product to our 1M+ readers and 2M+ followers on socials. Get in touch.

Until next time — Zain & the Superhuman AI team