MIT’s SEAL Technique Lets AI Teach Itself, MiniMax Launches Public Agent & Surge AI Quietly Becomes the Anti-Scale

Sign Up here | Advertise with Us

Welcome Back, AI Geeks!

It’s time to stay updated with What’s Up in AI.

A tiny team outperforms the big players, a new agent you can actually use drops in China, and MIT teaches models to learn like humans. Meanwhile, Wix bets $80M that anyone, even without code, can build the next big app with just a vibe and a prompt.

In today’s issue:

🤯 MiniMax Launches Public Agent
🧠 Surge AI Quietly Becomes the Anti-Scale
📈 MIT’s SEAL Technique Lets AI Teach Itself
⚡ 5-4-3-2-1 hits you just can't miss
😂 Your AI-Powered Laugh

Read time: 5 mins

LATEST IN AI TODAY📎

🤝 MiniMax Launches Public Agent You Can Actually Use

• MiniMax, backed by Alibaba, just dropped its general-purpose AI agent for public use.

• You can create coding projects, slideshows, reports, and more with an autonomous AI teammate.

• It performs on par with China’s other top agent, Manus, but is much easier to try.

• Includes a “Discover” feed with shared user projects like SEO dashboards and virtual museum tours.

• Each project comes with reusable prompts, making it easy to remix and build on others’ work.

📊 Surge AI Quietly Becomes the Anti-Scale

• As Meta buys 49% of Scale AI, rivals like OpenAI and Google are backing out.

• Enter Surge AI, a bootstrapped 110-person team that earned $1B last year — more than Scale.

• Like Scale, Surge focuses on high-quality data labeling for LLM training.

• The startup is getting attention as the next big supplier for frontier AI models.

• It’s part of a growing wave of lean, profitable AI infrastructure companies stepping into the spotlight.

🧠 MIT’s SEAL Technique Lets AI Teach Itself

• MIT built a method called SEAL (Self-Adapting Language Models) that lets AI tweak its own brain.

• Unlike static LLMs, SEAL enables real-time learning without retraining from scratch.

• Applied to Meta’s Llama and Alibaba’s Qwen, the method boosted performance noticeably.

• But too much adaptation can cause the models to “forget” previous knowledge.

• SEAL could bring us closer to truly adaptive, lifelong-learning AI.

BRAIN BOOST 🧠 

QUICK HITS 💥 

It’s time for some quick hits that you’ll love. Let’s go in…..4-3-2-1 👇️ 

5- TOP AI TOOLS OF THE DAY

🧑‍💼 Brainner
Screen hundreds of resumes fast with AI-powered reports that highlight top talent

📚 AI Manga Translator
Translate manga into 100+ languages while preserving original art and layout

🔗 Findr
Bring together AI and your scattered knowledge—apps, docs, links, notes, even meetings

📈 LiftKit
Boost conversions with ready-to-use GPT-4o marketing playbooks used by elite teams

🎨 Smart Palette
Turn your vibe into frame-worthy art—no design skills or complex prompts needed

4-AI JOB OPPORTUNITIES

• Harvey - Product Operations Manager

• xAI - Software Engineer, Ads

• Shield AI - Operations Officer, Flight Test

• Grammarly - Customer Success Manager

Want us to search for tech related jobs in your country/city ? Post a request 🔍

3-THINGS IN TREND

1. This mega-thread of X users sharing their coolest model context protocol (MCP) is a goldmine of good ideas

2. Are all LLMs converging around the same personality and responses? OpenAI cofounder Andrej Karpathy is on the case.

3. Duolingo built a full chess course from scratch with just two employees and the AI coding tool Cursor.

2-USEFUL LINKS

1. How AI chatbots keep you chatting!

2. How language models perform in human-like social scenarios

1-TOP ROBO NEWS

Meta’s New AI Model Helps Robots Understand the Real World 🧠🚗

Meta just unveiled V-JEPA 2, a powerful new AI model built to help robots and self-driving cars make sense of the physical world. Unlike traditional language models, V-JEPA 2 creates internal simulations of reality, letting machines reason about object movement, interactions, and real-world consequences with human-like logic. It’s a major step in Meta’s broader AI ambitions—and a big win for smarter, more capable robots.

WHAT ELSE HAPPENED IN AI 📰

🔻 OpenAI + Scale AI? Not So Fast
OpenAI is reportedly dialing back its work with Scale AI after the startup partnered with Meta — distancing itself like Google, xAI, and Microsoft already have.

🎥 Perplexity’s Video Power-Up
Perplexity now lets users create Veo 3 videos with audio by tagging @AskPerplexity on social media — bringing slick video gen straight to your feed.

🎙️ ChatGPT Can Now Record You
OpenAI launched ChatGPT Record, letting the assistant transcribe, summarize, and capture notes from your meetings and brainstorms.

🧬 Synthetic Science: SAIR Drops
SandboxAQ just released SAIR, a massive 5.2M synthetic protein-drug dataset backed by Nvidia to train next-gen drug discovery models.

❤️ AI That Sees Heart Risk
Researchers at Mass General Brigham created AI-CAC, a tool that scans chest CTs to detect calcium buildup linked to heart disease — way faster than traditional methods.

Funding 💰️ 

• Nabla Lands $70M To Build AI Agents In Healthcare Settings

GET AN AI-POWERED LAUGH 😁 

Acquire new customers and drive revenue by partnering with us

What’s Up in AI is one of the most engaged & loved newsletter for professionals, AI enthusiasts & business owners with ~45,000+ working at the world’s leading startups and enterprises. Companies like Hubspot, Artisan, Guidde & Gamma etc. feature their products in What’s Up in AI. You can learn more about partnering with us here.

And that’s a wrap!

Do drop a comment to tell us👇️ 

We’ll be back soon with more spicy takes on What’s Happening in AI, so stay tuned & share our newsletter with a friend using this link!

Stay Tuned! 🤖 
Team What’s Up in AI