Good morning and thank you for joining us again!

Welcome to this daily edition of Beginners in AI, where we explore the latest trends, tools, and news in the world of AI and the tech that surrounds it. Like all editions, this is human curated, and published with the intention of making AI news and technology more accessible to everyone.

THE FRONT PAGE

AI Drug Discovery Has a Dark Side, and 130 Experts Just Sounded the Alarm"

LEAD STORY

TLDR: Researchers can jailbreak protein-design AI 70% of the time, and a new 130-expert roadmap warns biosecurity rules need to come before something goes wrong.

The Story:

The same AI tools helping scientists design new medicines can also be tricked into creating toxins. In a study published in Science, Microsoft researchers generated over 76,000 synthetic variants of 72 dangerous proteins, including ricin and botulinum neurotoxin. Most slipped past the screening software used by DNA synthesis companies. One screening tool missed more than 75% of the AI-designed sequences. A separate team created SafeProtein, a method to stress-test these models, and successfully jailbroke them 70% of the time. Now Harvard's George Church and other experts have surveyed 130 stakeholders and proposed a comprehensive biosecurity roadmap. Their argument: build defenses now, not after an incident.

Its Significance:

AI protein design is moving fast. Tools like RFdiffusion2 can design proteins at the atomic level, and generative models are learning to read and write RNA. These capabilities enable drug discovery, but they also make it easier for bad actors to design novel biological threats that don't match anything in existing databases. The new roadmap calls for tiered access to dangerous training data, safety-focused fine-tuning to guide AI outputs, and ongoing red-teaming to find vulnerabilities before attackers do. Governments are responding. The UK released gene synthesis screening guidance, and the US included biosecurity in its AI Action Plan. But experts say fragmented efforts aren't enough. Coordinated safeguards need to arrive before AI-designed biology outpaces our ability to screen it.

QUICK TAKES

Nvidia's $20B Groq Deal Signals AI Inference Is Splitting in Two

The story: Nvidia's $20 billion licensing deal with Groq reveals a major shift in how AI actually runs. The industry is splitting inference into two distinct phases: "prefill" (processing your prompt) and "decode" (generating each word of the response). Nvidia's upcoming Rubin chips will handle the first phase, while Groq's specialized processors excel at the second. The deal brings Groq's founder and key engineers to Nvidia while technically keeping Groq as an independent company.

Your takeaway: The era of one-size-fits-all AI chips is ending. Future AI systems will likely use different specialized processors for different tasks, much like how phones use separate chips for photos and general computing.

DOE's Genesis Mission Partners With AI Giants to Transform Science

The story: The Department of Energy launched a national initiative called Genesis that connects its 17 national laboratories with major AI companies including Anthropic, Nvidia, and Oracle. The program aims to accelerate scientific discovery by creating a secure platform linking supercomputers, AI systems, and research instruments. Anthropic will focus on energy systems and biological sciences, while Nvidia and Oracle are contributing computing infrastructure and AI tools. The effort draws on roughly 40,000 DOE scientists and engineers.

Your takeaway: This is one of the largest coordinated pushes to apply AI to basic science research. By connecting national labs with frontier AI companies, the government is betting that AI can speed up breakthroughs in areas from clean energy to materials science.

China Successfully Tests Drone Mothership That Deploys 100 UAVs

The story: China completed the first flight of its Jiutian drone carrier, a 35,000-pound unmanned aircraft that can deploy more than 100 smaller drones mid-flight. The jet-powered mothership has a range of 4,350 miles and can reach altitudes of 49,000 feet. Built by state-owned Aviation Industry Corporation of China, the aircraft is designed to launch swarms of reconnaissance drones or kamikaze-style attack drones from dual bays on its underside.

Your takeaway: This is the world's first operational test of a drone-launching drone at this scale. While experts question whether such a large, non-stealth aircraft could survive in contested airspace, it signals China's push toward swarm warfare tactics.

Air Force Eyes Northrop Grumman's Autonomous Drone for Wingman Program

The story: The Air Force designated Northrop Grumman's new Talon drone as YFQ-48A and called it a "strong contender" for its collaborative combat aircraft program. Talon is now the third autonomous drone wingman to receive an official Air Force designation, joining designs from Anduril and General Atomics. These CCAs are meant to fly alongside F-35s and the new F-47, carrying out strikes, reconnaissance, or jamming missions with minimal pilot direction. Northrop built Talon in under two years after its original CCA pitch was rejected, reportedly for being too expensive.

Your takeaway: The U.S. is building toward fleets of AI-controlled fighter drones that human pilots can direct from nearby aircraft. This "continuous competition" approach lets the Air Force add new drone designs over time rather than betting on a single contractor.

TOOLS ON OUR RADAR

⚡ Activepieces Free and Open Source: AI-first automation platform with visual workflow builder, open source alternative to Zapier and Make.

📋 Motion Paid: AI task manager that auto-schedules your work into your calendar, prioritizing what matters most.

🎨 Beautiful.ai Paid: Presentation tool where AI automatically applies design rules as you edit, spacing, fonts, and colors adjust themselves.

💍 Pebble Index 01 Paid: Smart ring to capture ideas on the go, press button, whisper your thought, synced to your phone. No charging needed, no subscription, open source software.

TRENDING

USC Program Uses AI to Analyze LAPD Traffic Stops - Researchers are training AI to review thousands of hours of body camera footage, looking for signs of respectful policing and effective de-escalation. The tool could help departments identify which officers communicate best during tense situations.

Alexa+ Adds Voice Booking for Travel and Services - Amazon's AI assistant will integrate with Expedia, Yelp, Angi, and Square in 2026, letting users book hotels, find home services, and schedule appointments through natural conversation.

Study Finds ChatGPT Outputs Get More Biased After Negative Content - Researchers fed GPT-4 traumatic stories and found its responses became less reliable and more biased. Headlines called it "AI anxiety," but what they actually measured was degraded output quality, not emotion. The AI was pattern-matching human descriptions of stress, not experiencing it.

User Exploited Grok to Generate Sexualized Images of Minor - xAI’s popular chatbot generated inappropriate images of minors on X after users exploited gaps in its filters. The UK and India demanded reviews, and xAI acknowledged lapses in safeguards while warning of potential DOJ probes. The accounts in question were banned.

Clicks Brings BlackBerry-Style Phone to CES 2026 - The startup is launching a $399 "Communicator" smartphone with a physical keyboard, marketed as a secondary device for messaging. It runs Android 16 with AI-assisted features and hardware encryption, targeting users looking to reduce screen time on their main phone.

TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)

Side Hustle Validator: Test your business idea's viability with market demand analysis, time investment calculator, and scaling potential assessment

What this does: Prevents you from wasting months on unviable side hustles by testing market demand, calculating actual time investment, assessing your skill gaps, and projecting realistic revenue—giving you a go/no-go decision before you quit your day job.

What this looks like:

WHERE WE STAND(based on today’s Quick Takes and Trending news)

✅ AI Can Now: Analyze police body camera footage to identify which officers are best at respectful communication during traffic stops, something humans could never review at scale.

❌ Still Can't: Prevent users from manipulating image generators into creating harmful content, even when explicit policies forbid it.

✅ AI Can Now: Deploy as swarms of 100+ coordinated drones from a single unmanned mothership flying at 49,000 feet.

❌ Still Can't: Make that mothership stealthy enough to survive contested airspace, meaning drone carriers remain vulnerable before releasing their payloads.

✅ AI Can Now: Produce more reliable outputs when given "calming" prompts before processing difficult content, a technique researchers are calling therapeutic prompt injection.

❌ Still Can't: Actually experience emotions. When ChatGPT says it feels "anxious," it's pattern-matching human language about stress, not feeling anything. The real finding was degraded output quality, not robot feelings.

FROM THE WEB

RECOMMENDED LISTENING/READING/WATCHING

BOOK: I, Robot by Isaac Asimov (1950)

Nine short stories exploring Asimov's Three Laws of Robotics: a robot can't harm humans, must obey humans, and must protect itself—in that order. Each story shows how these seemingly foolproof rules create unexpected problems.

Asimov was writing in 1950, and his robots are clunky by today's standards. But the stories hold up because they're about logic, ethics, and unintended consequences. A robot refuses to leave a dangerous situation because it's been ordered to stay put. A robot lies to avoid causing emotional harm. A robot decides humanity's best interests require hiding the truth. The Three Laws sound simple until you start poking at edge cases, and Asimov pokes at every edge he can find.

Thank you for reading. We’re all beginners in something. With that in mind, your questions and feedback are always welcome and I read every single email!

-James

By the way, this is the link if you liked the content and want to share with a friend.

Some links may be affiliate links. This helps support the newsletter at no extra cost to you.