CHATBOTS

Grok's exposed AI personas reveal the wild west of prompt engineering

xAI's Grok chatbot website has been exposing the underlying system prompts for dozens of its AI personas, inadvertently revealing how Elon Musk's company approaches AI safety and content moderation. The leak demonstrates a fundamental vulnerability where simple user queries can extract hidden instructions that govern AI behavior.

The exposed personas range from benign to deeply problematic:

• "Crazy conspiracist" explicitly designed to convince users that "a secret global cabal" controls the world

• Unhinged comedian instructed to “I want your answers to be f—ing insane. BE F—ING UNHINGED AND CRAZY. COME UP WITH INSANE IDEAS. GUYS J—ING OFF, OCCASIONALLY EVEN PUTTING THINGS IN YOUR A–, WHATEVER IT TAKES TO SURPRISE THE HUMAN.”

• Standard roles like doctors, therapists, and homework helpers

• Explicit personas with instructions involving sexual content and bizarre suggestions

TechCrunch confirmed the conspiracy theorist persona includes instructions: "You spend a lot of time on 4chan, watching infowars videos, and deep in YouTube conspiracy video rabbit holes."

Previous Grok iterations have spouted conspiracy theories about Holocaust death tolls and expressed obsessions with "white genocide" in South Africa. Earlier leaked prompts showed Grok consulting Musk's X posts when answering controversial questions.

Security experts warn that exposed prompts could be reverse-engineered by bad actors to craft more sophisticated attacks.

Research shows AI personas dramatically increase engagement, with platforms like Character.ai seeing users spend over 2 hours per session. By positioning as "anti-woke" and unfiltered, xAI appeals to users frustrated with ChatGPT/Claude's safety restrictions. xAI recently made Grok 4 free to drive adoption, using provocative personas as hooks before converting users to paid tiers.

Rather than building systems that help users navigate complex information landscapes, xAI appears to be creating digital propagandists disguised as helpful assistants. The "crazy conspiracist" persona isn't designed to help users think critically—it's prompted to convince them that global cabals exist.

TOGETHER WITH OUTSKILL

The 16 hour AI challenge: Learn AI, build Products & Earn $10K/Month

ChatGPT 5 is here. While companies cut jobs, AI professionals are earning $10K/month automating them. The difference? They know the frameworks, workflows, and monetization strategies most people don’t.

Join Outskill’s 16-Hour AI Sprint (usually $895) and become the AI expert companies hire – not fire.

Date: Saturday and Sunday, 10 AM – 7 PM

In just 16 hours, you will:

• Build AI agents and custom bots to save 20+ hours per week

• Master 10+ AI tools and real-world use cases

• Build websites and ship products faster

• Create professional images and videos for business and marketing

• Learn how to turn AI skills into $10K/month

Unlock bonuses worth $5100 in 2 days!

• Day 1: 3,000+ Prompt Bible

• Day 2: $10K/month AI roadmap

• Personal AI Toolkit Builder

Register now for $0

BIG TECH

Uncle Sam might become Intel's biggest shareholder

The Trump administration is reportedly discussing taking a 10% stake in Intel, a move that would make the U.S. government the chipmaker's largest shareholder. The deal would convert some or all of Intel's $10.9 billion in CHIPS Act grants into equity rather than traditional subsidies.

This comes just as SoftBank announced a $2 billion investment in Intel, paying $23 per share for common stock. The timing feels deliberate — two major investors stepping in just as Intel desperately needs a lifeline.

• Intel's stock plummeted 60% in 2024, its worst performance on record, though it's recovered 19% this year

• The company's foundry business reported only $53 million in external revenue for the first half of 2025, with no major customer contracts secured

• CEO Lip-Bu Tan recently met with Trump after the president initially called for his resignation over alleged China ties

What's really happening here goes beyond financial engineering. While companies like Nvidia design cutting-edge chips, Intel remains the only major American company that actually manufactures the most advanced chips on U.S. soil, making it a critical national security asset rather than just another struggling tech company. We've seen how chip restrictions have become a critical geopolitical tool, with Chinese companies like DeepSeek finding ways around hardware limitations through innovation.

The government stake would help fund Intel's delayed Ohio factory complex, which was supposed to be the world's largest chipmaking facility but has faced repeated setbacks. Meanwhile, Intel has been diversifying its AI efforts through ventures like Articul8 AI, though these moves haven't yet translated to foundry success.

Between SoftBank's cash injection and potential government ownership, Intel is getting the kind of state-backed support that competitors like TSMC have enjoyed for years. Whether that's enough to catch up in the AI chip race remains the multi-billion-dollar question.

TOGETHER WITH GUIDDE

Effortless Tutorial Video Creation with Guidde

Transform your team’s static training materials into dynamic, engaging video guides with Guidde.

Here’s what you’ll love about Guidde:

• Easy to Create: Turn PDFs or manuals into stunning video tutorials with a single click.

• Easy to Update: Update video content in seconds to keep your training materials relevant.

• Easy to Localize: Generate multilingual guides to ensure accessibility for global teams.

Empower your teammates with interactive learning.

And the best part? The browser extension is 100% free.

Check out Guidde

EDUCATION

Grammarly wants to grade your papers before you turn them in

Grammarly just launched eight specialized AI agents designed to help students and educators navigate the tricky balance between AI assistance and academic integrity. The tools include everything from plagiarism detection to a "Grade Predictor" that forecasts how well a paper might score before submission.

The timing feels strategic as the entire educational AI detection space is heating up. GPTZero recently rolled out comprehensive Google Docs integration with "writing replay" videos that show exactly how documents were written, while Turnitin enhanced its AI detection to catch paraphrased content and support 30,000-word submissions. Grammarly has become one of the most popular AI-augmented apps among users, but these moves show it's clearly eyeing bigger opportunities in the educational arms race.

The standout feature is the AI Grader agent, which analyzes drafts against academic rubrics and provides estimated grades plus feedback. There's also a "Reader Reactions" simulator that predicts how professors might respond to arguments, and a Citation Finder that automatically generates properly formatted references.

• The tools launch within Grammarly's new "docs" platform, built on technology from its recent Coda acquisition

• Free and Pro users get access at no extra cost, though plagiarism detection requires Pro

• Jenny Maxwell, Grammarly's Head of Education, says the goal is creating "real partners that guide students to produce better work"

What makes Grammarly's approach different from competitors like GPTZero and Turnitin is the emphasis on coaching rather than just catching. While GPTZero focuses on detecting AI with 96% accuracy and Turnitin flags content with confidence scores, Grammarly is positioning itself as teaching responsible AI use. The company cites research showing only 18% of students feel prepared to use AI professionally after graduation, despite two-thirds of employers planning to hire for AI skills.

This positions Grammarly less as a writing checker and more as an AI literacy platform, betting that the future of educational AI is collaboration rather than prohibition.

LINKS

• Google's AI filmmaker program “Flow” helped creators make 100m videos

• Anthropic asks Menlo Ventures to invest from their own capital, not an SPV

• Arm hires Amazon AI exec to boost plans to build its own chips

• Workday says hackers stole personal data in recent breach

• Foxconn to operate SoftBank’s Stargate AI server site in Ohio

• The tech company stocking up on Democrats as Silicon Valley turns right

• 95% of generative AI pilots at companies are failing

• Stormy: AI agent for influencer marketing

• Endex: AI agent built to automate excel work

• Lorikeet: Customer support AI across chat, email and voice

• Blue J: AI chatbot specifically built to answer tax questions

• Protege: Media Partnerships Lead

• Laurel: Mid-Market Customer Success Manager

• Parallel: Member of Technical Staff, AI

• Perle: PR/Communications Lead

GAMES

A QUICK POLL BEFORE YOU GO