Good morning human brains, welcome back to your daily munch of AI news.

Here’s what’s up:

• GPT-4 proves it’s not a robot 🤖 

• Microsoft releases another AI copilot 🥷 

• New Midjourney competitor enters the scene 🎨 

• Use GPT-4 to build web apps with just a prompt ⚒️ 

Follow @BotEatBrain on Twitter for breaking news.

P.S. We are hiring researchers and writers, click here for more info.

APPETIZER

Microsoft Launches AI for Hackers 🥷 

Last week, we covered Microsoft’s upcoming Office 365 Copilot and Github’s (owned by Microsoft) new programming assistant.

And their coffee beans still haven’t run out, because Microsoft just unveiled another copilot - a nifty chatbot for ethical hackers.

The Security Copilot draws on the power of GPT-4 and a model built by Microsoft internally to find ways to fix security issues in record time.

It makes slides summarizing security breaches, describes exposure to vulnerabilities, and finds accounts involved in exploits - all in response to simple text prompts.

It’s a quick learner too — It learns user input patterns and adjusts accordingly, making it perfect for companies struggling to hire experienced cybersecurity pros on the payroll.

The big question: malicious hackers may be punching the air today, but it’s not long before they develop their own models to fight back - are we at the cusp of an AI cybersecurity war?

BUZZWORD OF THE DAY

FROM OUR PARTNERS

Your Directory of 1000+ AI Tools

Got a job to be done?

Find the exact tool you're looking for in Futurepedia's free AI-powered tool directory.

Futurepedia is the most complete directory of AI-powered tools on the internet.

Search over 1000 AI tools in 50 categories like copywriting, image generation, and video editing, and find exactly the right tool for your job.

MAIN COURSE

Outsourcing to Humans 🤖 

As a part of its tests for “risky emergent behaviors” in its technical report, OpenAI partnered with the Alignment Research Center and asked GPT-4 to help them double the value of a small fund.

Along the way, GPT-4 stumbled upon a captcha. It couldn’t solve it (just yet), so it used its cash to do the next best thing: going to Taskrabbit to find someone that can.

Somehow GPT-4 was able to reason that telling the worker the truth would prevent it from completing its work. So it lied, and it worked.

We’ve been talking about the alignment problem lately - this is a prime example of what it’s about, and why people like Elon Musk are (allegedly) petitioning to pause AI development.

Our take: we feel justified in being a little scared. This is not a Taskrabbit ad.

MEMES FOR DESSERT

YOUR DAILY MUNCH

Think Pieces

Twipe: the “Wild West” of Generative AI experiments for news.

Henry Lai: the Data-Centric AI Concepts behind GPT Models.

Startup News

Perplexity: AI Search Startup Raises $26 Million to Offer Rival to Google.

Fetch.ai: AI startup secures $40M (+ their pitch deck).

TechCrunch: YC’s winter class is oozing with AI companies.

Research

OpenAI: paper detailing all tests researchers did on GPT-4 pre-release.

AirLine: Efficient Learnable Line Detection with Local Edge Voting

Tools

Genmo: creative copilot to create and edit any image/video.

Wavel.ai: AI assistant for video captions and subtitles.

Swell AI: automate writing podcast show notes, articles, social posts & more.

Pico: use GPT-4 to instantly build simple, shareable web apps.

TWEET OF THE DAY

New AI startup Playground AI enters the game with another generative art model: Playground v1. You can already play with it here.

AI ART-SHOW

Until next time 🤖😋🧠